This is a revised version of the UNIX compendium which is available in printed form and online via the WWW and info hypertext readers. It forms the basis for a one or two semester course in UNIX. The most up-to-date version of this manual can be found at
http://www.iu.hio.no/~mark/unix/unix.html.
It is a reference guide which contains enough to help you to find what you need from other sources. It is not (and probably can never be) a complete and self-contained work. Certain topics are covered in more detail than others. Some topics are included for future reference and are not intended to be part of an introductory course, but will probably be useful later. The chapter on X11 programming has been deleted for the time being.
Comments to Mark.Burgess@iu.hio.no Oslo, August 2001
If you are coming to unix for the first time, from a Windows or MacIntosh environment, be prepared for a rather different culture than the one you are used to. UNIX is not about `products' and off-the-shelf software, it is about open standards, free software and the ability to change just about everything.
You should approach UNIX the way you should approach any new system: with an open mind. The journey begins...
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".
UNIX is one of the most important operating system in use today, perhaps even the most important. Since its invention around the beginning of the 1970s it has been an object of continual research and development. UNIX is not popular because it is the best operating system one could imagine, but because it is an extremely flexible system which is easy to extend and modify. It is an ideal platform for developing new ideas.
Much of the success of UNIX may be attributed to the rapid pace of its
development (a development to which all of its users have been able to
contribute) its efficiency at running programs and the many powerful
tools which have been written for it over the years, such as the C
programming language, make
, shell, lex
and yacc
and
many others. UNIX was written by programmers for programmers. It is
popular in situations where a lot of computing power is required and for
database applications, where timesharing is critical. In contrast to
some operating systems, UNIX performs equally well on large scale
computers (with many processors) and small computers which fit in your
suitcase!
All of the basic mechanisms required of a multi-user operating system are present in UNIX. During the last few years it has become ever more popular and has formed the basis of newer, though less mature, systems like NT. One reason for this that now computers have now become powerful enough to run UNIX effectively. UNIX places burdens on the resources of a computer, since it expects to be able to run potentially many programs simultaneously.
If you are coming to UNIX from Windows or DOS you may well be used to using applications software or helpful interactive utilities to solve every problem. UNIX is not usually like this: the operating system has much greater functionality and provides the possibilities for making your own, so it is less common to find applications software which implements the same things. In UNIX you are usually asked to learn a language in order to express exactly what you want. This is much more powerful than menu systems, but it is harder to learn
UNIX has long been in the hands of academics who are used to making their own applications or writing their own programs, whereas as the Windows world has been driven by businesses who are willing to spend money on software. For that reason commercial UNIX software is often very expensive and therefore not available at this college. On the other hand, the flexibility of UNIX means that it is easy to write programs and it is possible to fetch gigabytes of free software from the Internet to suit your needs. It may not look exactly like what you are used to on your PC, but then you have to remember that UNIX users are a different kind of animal altogether
Like all operating systems, UNIX has many faults. The biggest problem for any operating system is that it evolves without being redesigned. Operating systems evolve as more and more patches and hacks are applied to solve day-to-day problems. The result is either a mess which works somehow (like UNIX) or a blank refusal to change (like DOS or the MacIntosh, prior to MacOS X, which is based on BSD UNIX). From a practical perspective, UNIX is important and successful because it is a multi-process system which
UNIX has some problems: it is old, it contains a lot of rubbish which no one ever bothered to throw away. Although it develops quickly (at light speed compared to either DOS/Windows or MacIntosh) the user interface has been the slowest thing to change. UNIX is not user-friendly for beginners, it is user-friendly for advanced users: it is made for users who know about computing. It sometimes makes simple things difficult, but above all it makes things possible!
The aim of this introduction is to
To accomplish this task, we must first learn something about the shell language (the way in which UNIX starts programs). Later we shall learn how to solve more complex problems using Perl and C. Each of these is a language which can be used to put UNIX to work. We must also learn when to use which tool, so that we do not waste time and effort. Typical uses for these different interfaces are
Much of UNIX's recent popularity has been a result of its networking abilities: UNIX is the backbone of the Internet. No other widely available system could keep the Internet alive today. GNU/Linux is a free/open source re-write of the UNIX operating system, which many enhancements. While GNU/Linux is not "rocket science" to computer experts, it has distilled the essence of UNIX and placed it in the hands of everyone. It runs on wrist watches and mainframe computers. Like it or loathe it, GNU/Linux is probably the most important single development in computer operating systems for many years.
Once you have mastered the UNIX interface and philosophy you will find that i) the PC and MacIntosh window environments might seem to be easy to use, but are simplistic and primitive by comparison; ii) UNIX is far from being the perfect operating system--it has a whole different set of problems and flaws.
The operating system of the future will not be UNIX or GNU/Linux as we see it today (hopefully), nor will is be DOS or MacIntosh, but one thing is for certain: it will owe a lot to the UNIX operating system and will contain many of the tools and mechanisms we shall describe below.
UNIX is not a single operating system. It has branched out in many different directions since it was introduced by AT&T. The most important `fork()' in its history happened early on when the university of Berkeley, California created the BSD (Berkeley Software Distribution), adding network support and the C-shell.
Here are some of the most common implementations of unix.
This programming guide is something between a user manual and a tutorial. The information contained here should be sufficient to get you started with the unix system, but it is far from complete.
To use this programming guide, you will need to work through the basics from each chapter. You will find that there is much more information here than you need straight away, so try not to be overwhelmed by the amount of material. Use the contents and the indices at the back to find the information you need. If you are following a one-semester UNIX course, you should probably concentrate on the following:
The only way to learn UNIX is to sit down and try it. As with any new thing, it is a pain to get started, but once you are started, you will probably come to agree that UNIX contains a wealth of possibilities, perhaps more than you had ever though was possible or useful!
One of the advantages of the UNIX system is that the entire UNIX manual is available on-line. You should get used to looking for information in the online manual pages. For instance, suppose you do not remember how to create a new directory, you could do the following:
nexus% man -k dir dir ls (1) - list contents of directories dirname dirname (1) - strip non-directory suffix from file name dirs bash (1) - bash built-in commands, see bash(1) find find (1) - search for files in a directory hierarchy ls ls (1) - list contents of directories mkdir mkdir (1) - make directories pwd pwd (1) - print name of current/working directory rmdir rmdir (1) - remove empty directories
The `man -k' command looks for a keyword in the manual and lists all the references it finds. The command `apropos' is completely equivalent to `man -k'. Having discovered that the command to create a directory is `mkdir' you can now look up the specific manual page on `mkdir' to find out how to use it:
man mkdir
Some but no all of the UNIX commands also have a help option which is activated with the `-h' or `--help' command-line option.
dax% mkdir --help Usage: mkdir [OPTION] DIRECTORY... -p, --parents no error if existing, make parent directories as needed -m, --mode=MODE set permission mode (as in chmod), not 0777 - umask --help display this help and exit --version output version information and exit dax%
There are some things that you should never do in UNIX. Some of these will cause you more serious problems than others. You can make your own list as you discover more.
rm
it is impossible
to recover it! Don't use wildcards with rm
without thinking
quite carefully about what you are doing! It has happened to very many
users throughout the history of UNIX that one tries to type
rm *~but instead, by a slip of the hand, one writes
rm * ~UNIX then takes these wildcards in turn, so that the first command is
rm *
which deletes all of your files! BE CAREFUL!
test
. There is a UNIX command
which is already called test and chances are that when you try to
run your program you will start the UNIX command instead. This
can cause a lot of confusion because the UNIX command doesn't seem
to do very much at all!
The core of unix is the library of functions (written in C) which access the system. Everything you do on a unix system goes through this set of functions. However, you can choose your own interface to these library functions. UNIX has very many different interfaces to its libraries in the form of languages and command interpreters.
You can use the functions directly in C, or you can use command programs like `ls', `cd' etc. These functions just provide a simple user interface to the C calls. You can also use a variety of `script' languages: C-shell, Bourne shell, Perl, Tcl, scheme. You choose the interface which solves your problem most easily.
With the exception of a few simple commands which are built into the command interpreter (shell), all unix commands and programs consist of executable files. In other words, there is a separate executable file for each command. This makes it extremely simple to add new commands to the system. One simply makes a program with the desired name and places it in the appropriate directory.
UNIX commands live in special directories
(usually called bin
for binary files). The location of
these directories is recorded in a variable called path
or
PATH
which is used by the system to search for
binaries. We shall return to this in more detail in later chapters.
Since users cannot command the kernel directly, UNIX has a command language known as the shell. The word shell implies a layer around the kernel. A shell is a user interface, or command interpreter.
There are two main versions of the shell, plus a number of enhancements.
The program tcsh
is a public-domain enhancement of the csh and is
in common use. Two improved versions of the Bourne shell also exist:
ksh
, the Korn shell and bash
, the Bourne-again shell.
Although the shells are mainly tools for typing in commands (which are executable files to be loaded and run), they contain features such as aliases, a command history, wildcard-expansions and job control functions which provide a comfortable user environment.
Most of the unix kernel and daemons are written in the C programming
language (1). Calls to the kernel and to
services are made through functions in the standard C library. The
commands like chmod
, mkdir
and cd
are all C
functions. The binary files of the same name /bin/chmod
,
/bin/mkdir
etc. are just trivial "wrapper" programs for these C
functions.
Until Solaris 2, the C compiler was a standard part of the UNIX operating system, thus C is the most natural language to program in in a UNIX environment. Some tools are provided for C programmers:
UNIX has three logical streams or files which are always open and are available to any program.
The names are a part of the C language and are defined as
pointers of type FILE
.
#include <stdio.h> /* FILE *stdin, *stdout, *stderr; */ fprintf(stderr,"This is an error message!\n");
The names are `logical' in the sense that they do not refer to a particular device, or a particular place for information to come from or go. Their role is analogous to the `.' and `..' directories in the filesystem. Programs can write to these files without worrying about where the information comes from or goes to. The user can personally define these places by redirecting standard I/O. This is discussed in the next chapter.
A separate stream is kept for error messages so that error output does not get mixed up with a program's intended output.
When logged onto a UNIX system directly, the user whose name is
root
has unlimited access to the files on the system. root
can also become any other user without having to give a
password. root
is reserved for the system administrator or
trusted users.
Certain commands are forbidden to normal users. For example, a regular
user should not be able to halt the system, or change the ownership
of files (see next paragraph). These things are reserved for the
root
or superuser.
In a networked environment, root
has no automatic authority on
remote machines. This is to prevent the system administrator of one
machine in Canada from being able to edit files on another in China. He
or she must log in directly and supply a password in order to gain
access privileges. On a network where files are often accessible in
principle to anyone, the username root
gets mapped to the user
nobody
, who has no rights at all.
UNIX has a hierarchical filesystem, which makes use of directories and sub-directories to form a tree. The root of the tree is called the root filesystem or `/'. Although the details of where every file is located differ for different versions of unix, some basic features are the same. The main sub-directories of the root directory together with the most important file are shown in the figure. Their contents are as follows.
mknod
. Logical
devices are UNIX's official entry points for writing to devices. For instance,
/dev/console
is a route to the system console, while /dev/kmem
is a route for reading kernel memory. Device nodes enable devices to be
treated as though they were files.
/home
by some convention decided by the system administrator.
/usr/spool
contains spool queues and system
data. /var/spool
and /var/adm
etc are used for holding
queues and system log files.
Every unix directory contains two `virtual' directories marked by a single dot and two dots.
ls -a . ..
The single dot represents the directory one is already in (the current directory). The double dots mean the directory one level up the tree from the current location. Thus, if one writes
cd /usr/local cd ..
the final directory is /usr
. The single dot is very useful in
C programming if one wishes to read `the current directory'. Since
this is always called `.' there is no need to keep track of what the
current directory really is.
`.' and `..' are `hard links' to the true directories.
A symbolic link is a pointer or an alias to another file. The command
ln -s fromfile /other/directory/tolink
makes the file fromfile
appear to exist at /other/directory/tolink
simultaneously. The file is not copied, it merely appears to be a
part of the file tree in two places. Symbolic links can be made to
both files and directories.
A symbolic link is just a small file which contains the name of the real
file one is interested in. It cannot be opened like an ordinary file,
but may be read with the C call readlink()
See section lstat and readlink.
If we remove the file a symbolic link
points to, the link remains -- it just points nowhere.
A hard link is a duplicate inode in the filesystem which is in every way equivalent to the original file inode. If a file is pointed to by a hard link, it cannot be removed until the link is removed. If a file has @math{n} hard links -- all of them must be removed before the file can be removed. The number of hard links to a file is stored in the filesystem index node for the file.
If you have never met unix, or another multiuser system before, then you might find the idea daunting. There are several things you should know.
Each time you use unix you must log on to the system by typing a username and a password. Your login name is sometimes called an `account' because some unix systems implement strict quotas for computer resources which have to be paid for with real money(2).
login: mark password:
Once you have typed in your password, you are `logged on'. What happens then depends on what kind of system you are logged onto and how. If you have a colour monitor and keyboard in front of you, with a graphical user interface, you will see a number of windows appear, perhaps a menu bar. You then use a mouse and keyboard just like any other system.
This is not the only way to log onto unix. You can also log in
remotely, from another machine, using the Secure Shell ssh
program ( this
replaces the now antiquated telnet
and rlogin
programs). If you use these programs, you will normally only get a text
or command line interface (though graphical interfaces can easily be arranged).
Once you have logged in, a short message will be printed (called Message of the Day or motd) and you will see the C-shell prompt: the name of the host you are logged onto followed by a percent sign, e.g.
Linux cube 2.2.19pre13 #2 Mon Feb 26 15:53:31 MET 2001 i686 unknown This is GNU/Linux - send problems to help@example.org 10:44pm up 8 days, 13:34, 3 users, load average: 0.08, 0.02, 0.01 There are 480 messages in your incoming mailbox.
Remember that every UNIX machine is a separate entity: it is not like logging onto a PC system where you log onto the `network' i.e. the PC file server. Every UNIX machine is a server, or a client -- more correctly a "peer" (equal partner). The network, in unix-land, has lots of players.
The first thing you should do once you have logged on is to set a reliable password. A poor password might be okay on a PC which is not attached to a large network, but once you are attached to the Internet, you have to remember that the whole world will be trying to crack your password. Don't think that no one will bother: some people really have nothing better to do. A password should not contain any word that could be in a list of words (in any language), or be a simple concatenation of a word and a number (e.g. mark123). It takes seconds to crack such a password. Choose instead something which is easy to remember. Feel free to use the PIN number from your bankers card in your password! This will leave you with fewer things to remember. e.g. Ma9876rk). Passwords can be up to eight characters long.
Some sites allow you to change your password anywhere. Other sites require you to log onto a special machine to change your password:
dax% dax% passwd Change your password on host nexus You cannot change it here dax% rlogin nexus password: ****** nexus% passwd Changing password for mark Enter login password: ******** Enter new password: ******** Reenter new passwd: ********
You will be prompted for your old password and your new password twice. If your network is large, it might take the system up to an hour or two to register the change in your password, so don't forget the old one right away!
UNIX has three mouse buttons. On some PC's running GNU/Linux or some other PC unix, there are only two, but the middle mouse button can be simulated by pressing both mouse buttons simultaneously. The mouse buttons have the following general functions. They may also have additional functions in special software.
On a left-handed system right and left are reversed.
Reading electronic mail on unix is just like any other system, but there are many programs to choose from. There are very old programs from the seventies such as
and there are fully graphical mail programs such as
tkrat mailtool
Choose the program you like best. Not all of the programs support modern
multimedia extensions because of their age. Some programs like
tkrat
have immediate mail notification alerts. To start a mail
program you just type its name. If you have an icon-bar, you can click
on the mail-icon.
Inexperienced computer users often prefer to use file-manager
programs to avoid typing anything. With a mouse you can click
your way through directories and files without having to type
anything (e.g. the kfm
or tkdesk
programs).
More experienced users generally find this to be slow and
tedious after a while and prefer to use written commands.
UNIX has many short cuts and keyboard features which make
typed commands extremely fast and much more powerful than
use of the mouse.
Today the CDE, KDE and GNOME projects are the most important efforts to write graphical user interfaces for computers. The CDE (Common Desktop Environment) is a commercial program developed by IBM, Hewlett-Packard, Sun Microsystems and many other vendors. KDE (a German effort, a pun on CDE) and GNOME are free software window systems which have taken windowing to the next level. While they have borrowed and stolen many ideas from Windows' innovative Windows 95 user interface, they have taken windowing beyond this.
If you come from a Windows environment, the UNIX commands can be a little strange. It is a different way of thinking: using language to ask for exactly what you want, instead of pointing to a menu of limited choices. It is also a strange language. Because they stem from an era when keyboards had to be hit with hammer force, and machines were very slow, the UNIX command names are as short as possible, so they seem pretty cryptic. Some familiar ones which DOS borrowed from UNIX include,
cd mkdir
which change to a new directory and make a new directory respectively. To list the files in the current directory you use,
ls
To rename a file, you `move' it:
mv old-name new-name
Text editing is one of the things which people spend most time doing on any computer. It is important to distinguish text editing from word processing. On a PC or MacIntosh, you are perhaps used to Word or WordPerfect for writing documents.
UNIX has a Word-like program called lyx
, and even several
Office clones (e.g. Star Office soffice
), but for the most part
UNIX users do not use word processors. It is more common in the UNIX
community to write all documents, regardless of whether they are
letters, books or computer programs, using a non-formatting text
editor. (UNIX word processors like Framemaker
do exist, but they
are very expensive. A version of MS-Word also exists for some unices.)
Once you have written a document in a normal text editor, you call up
a text formatter to make it pretty. You might think this strange, but
the truth of the matter is that this two-stage process gives you the
most power and flexibility--and that is what most UNIX folks like.
For writing programs, or anything else, you edit a file by typing:
emacs myfile
emacs
is one of dozens of text-editors. It is not the simplest or
most intuitive, but it is the most powerful and if you are going to
spend time learning an editor, it wouldn't do any harm to make it this
one. You could also click on emacs' icon if you are relying on a window
system. Emacs is almost certainly the most powerful text editor that
exists on any system. It is not a word-processor, it is not for formatting
printed documents, but it can be linked to almost any other program in
order to format and print text. It contains a powerful programming
language and has many intelligent features. We shall not go into the
details of document formatting in this book, but only mention that
programs like troff
and Tex
or Latex
are used for
this purpose to obtain typeset-quality printing. Text formatting is an
area where UNIX folks do things differently to PC folks.
UNIX began as a timesharing mainframe system in the seventies, when the only terminals available were text based teletype terminals or tty-s. Later, the Massachusetts Institute of Technology (MIT) developed the X-windows interface which is now a standard across UNIX platforms. Because of this history, the X-window system works as a front end to the standard UNIX shell and interface, so to understand the user environment we must first understand the shell.
A shell is a command interpreter. In the early days of UNIX, a shell was the only way of issuing commands to the system. Nowadays many window-based application programs provide menus and buttons to perform simple commands, but the UNIX shell remains the most powerful and flexible way of interacting with the system.
After logging in and entering a password, the UNIX process init starts a shell for the user logging in. UNIX has several different kinds of shell to choose from, so that each user can pick his/her favourite command interface. The type of shell which the system starts at login is determined by the user's entry in the passwd database. On most systems, the standard login shell is a variant of the C-shell.
Shells provide facilities and commands which
The shell does not contain any more specific functions--all other commands, such as programs which list files or create directories etc., are executable programs which are independent of the shell. When you type `ls', the shell looks for the executable file called `ls' in a special list of directories called the command path (which is contained in the environment variable $PATH) and attempts to start this program. This allows such programs to be developed and replaced independently of the actual command interpreter.
Each shell which is started can be customized and configured by editing a setup file. For the Bash shell this file is called `.bashrc', and for the C-shell and its variants it is called `.profile'. (Note that files which begin with leading dots are not normally visible with the `ls' command. Use `ls -a' to view these.) Any commands which are placed in these files are interpreted by the shell before the first command prompt is issued. These files are typically used to define a command search path and terminal characteristics.
On each new command line you can use the cursor keys to edit the line. The up-arrow browses back through earlier commands. CTRL-a takes you to the start of the line. CTRL-e takes you to the end of the line. The TAB can be used to save typing with the `completion' facility See section Command/filename completion.
Shell commands are commands like cp
, mv
,
passwd
, cat
, more
, less
, cc
,
grep
, ps
etc..
One thing you can always bet on with Unix is that there is not
just one way of doing things -- there are so many standards, that
there is often a bewildering array to choose from. UNIX has two main
command shells. They are called sh
(Bourne Shell) and csh
C-shell. Their
modern implementations are called Bash
(Bourne Again Shell)
and tcsh
(T-C shell).
Very few commands are actually built into the shell command line
interpreter, in the same way that they are built into DOS. Rather
commands are programs which exist as actual program files. When we type
a command, the shell searches for a program with the same name and tries
to execute it. This is very flexible, since anyone is free to write
their own programs and therefore extend the command language of the
system. The
file must be executable, or a Command not found
error will
result. To see what actually happens when you type a command like
gcc
, try typing the following into a GNU/Linux system: (you
can type this exactly as shown into a Bash shell)
cube$ IFS=: cube$ for dir in $PATH # for every directory in the list path >do > if [ -x $dir/gcc ] # if the file is executable > then > echo Found $dir/gcc # Print message found! > break # break out of loop > else > echo Searching $dir/gcc > fi >done
If you use C-shell (e.g. tcsh), try typing in the following C-shell commands directly into a C-shell.
nexus% foreach dir ( $path ) # for every directory in the list path > if ( -x $dir/gcc ) then # if the file is executable > echo Found $dir/gcc # Print message found! > break # break out of loop > else > echo Searching $dir/gcc > endif > end
The output of these command sequences is something like this:
Searching /usr/lang/gcc Searching /usr/openwin/bin/gcc Searching /usr/openwin/bin/xview/gcc Searching /physics/lib/framemaker/bin/gcc Searching /physics/motif/bin/gcc Searching /physics/mutils/bin/gcc Searching /physics/common/scripts/gcc Found /physics/bin/gcc
If you type
echo $PATH
in Bourne Shell, or
echo $path
in C-shell you will see the entire list of directories which are searched by the shell. If we had left out the `break' command, we might have discovered that UNIX often has several programs with the same name, in different directories! For example,
/bin/mail /usr/ucb/mail /bin/Mail /bin/make /usr/local/bin/make.
Also, different versions of UNIX have different conventions for placing the
commands in directories, so the path list needs to be different for
different types of UNIX machine. In Bash a few basic commands
like cd
and kill
are built into the shell (as in DOS).
You can find out which directory a command is stored in using
type
command. For example
cube$ type cd cd is a shell builtin cube$ type mv mv is /bin/mv cube$
type
only searches the directories in $PATH
and quits after
the first match, so if there are several commands with the same name,
you will only see the first of them using type
.
Finally, in the C-shell the command corresponding to type is built in and
called which
. In Bash which
is a program:
cube$ type which which is /usr/bin/which cube$ tcsh cube% which which which: shell built-in command.
Take a look at the script /usr/bin/which
. It is a script written
in bash.
Environment variables are variables which the shell keeps. They are
normally used to configure the behaviour of utility programs like
lpr
(which sends a file to the printer) and mail
(which
reads and sends mail) so that special options do not have to be typed in
every time you run these programs.
Any program can read these variables to find out how you have configured your working environment. We shall meet these variables frequently. Here are some important variables
PATH # The search path for shell commands (bash) TERM # The terminal type (bash and csh) DISPLAY # X11 - the name of your display LD_LIBRARY_PATH # Path to search for object and shared libraries HOSTNAME # Name of this UNIX host PRINTER # Default printer (lpr) HOME # The path to your home directory (bash) PS1 # The default prompt for bash path # The search path for shell commands (csh) term # The terminal type (csh) prompt # The default prompt for csh home # The path to your home directory (csh)
These variables fall into two groups. Traditionally the first group always have names in uppercase letters and are called environment variables, whereas variables in the second group have names with lowercase letters and are called shell variables-- but this is only a convention. The uppercase variables are global variables, whereas the lower case variables are local variables. Local variables are not defined for programs or sub-shells started by the current shell, while global variables are inherited by all sub-shells.
The Bash-shell and the C-shell use these conventions differently and not
always consistently. You will see how to define these below. For now
you just have to know that you can use the command env
can be used
in Bash shell to see all of the defined global environment variables while
set
lists both the global and the local variables.
Sometimes you want to be able to refer to several files in one go. For
instance, you might want to copy all files ending in `.c' to a new
directory. To do this one uses wildcards. Wildcards are characters
like * ?
which stand for any character or group of characters.
In card games the joker is a `wild card' which can be substituted for
any other card. Use of wildcards is also called filename substitution
in the UNIX manuals, in the sections on sh
and csh
.
The wildcard symbols are,
ls /etc/rc.????
ls /etc/rc.*
ls [abc].C
Here are some examples and explanations.
rc.
and are 7 characters long.
It is important to understand that the shell expands wildcards. When
you type a command, the program is not invoked with an argument that
contains *
or ?
. The shell expands the special characters
first and invokes commands with the entire list of files which
match the patterns. The programs never see the wildcard characters, only
the list of files they stand for. To see this in action, you can type
echo /etc/rc*
which gives
/etc/rc0 /etc/rc0.d /etc/rc1 /etc/rc1.d /etc/rc2 /etc/rc2.d /etc/rc3 /etc/rc3.d /etc/rc5 /etc/rc6 /etc/rcS /etc/rcS.d
All shell commands are invoked with a command line of this form. This has an important corollary. It means that multiple renaming cannot work!
UNIX files are renamed using the mv
command. In many microcomputer
operating systems one can write
rename *.x *.y
which changes the file extension of all files ending in `.x' to the same name with a `.y' extension. This cannot work in UNIX, because the shell tries expands everything before passing the arguments to the command line.
The wildcards belong to the shell. They are used for matching filenames. UNIX has a more general and widely used mechanism for matching strings, this is through regular expressions.
Regular expressions are used by the egrep
utility, text editors
like ed
, vi
and emacs
and sed
and awk
.
They are also used in the C programming language
for matching input as well as in the Perl programming language and lex
tokenizer. Here are some examples using the egrep
command
which print lines from the file /etc/rc
which match certain
conditions. The construction is part of
egrep
. Everything
in between these symbols is a regular expression. Notice that
special shell symbols ! * &
have to be preceded with a backslash
\
in order to prevent the shell from expanding them!
# Print all lines beginning with a comment # egrep '(^#)' /etc/rc # Print all lines which DON'T begin with # egrep '(^[^#])' /etc/rc # Print all lines beginning with e, f or g. egrep '(^[efg])' /etc/rc # Print all lines beginning with uppercase egrep '(^[A-Z])' /etc/rc # Print all lines NOT beginning with uppercase egrep '(^[^A-Z])' /etc/rc # Print all lines containing ! * & egrep '([\!\*\&])' /etc/rc # All lines containing ! * & but not starting # egrep '([^#][\!\*\&])' /etc/rc
Regular expressions are made up of the following `atoms'.
These examples assume that the file `/etc/rc' exists.
If it doesn't exist on the machine you are using, try to
find the equivalent by, for instance, replacing
/etc/rc
with /etc/rc*
which will try to
find a match beginning with the rc.
You can find a complete list in the UNIX manual pages. The square brackets above are used to define a class of characters to be matched. Here are some examples,
The backwards apostrophes `...` can be used in all shells and also in the programming language Perl. When these are encountered in a string the shell tries to execute the command inside the quotes and replace the quoted expression by the result of that command. For example:
UNIX$ echo "This system's kernel type is `/usr/bin/file /boot/vmlinuz-2.2.19pre13`" This system's kernel type is /boot/vmlinuz-2.2.19pre13: Linux kernel x86 boot executable bzImage, version 2.2.19pre13 UNIX$ for file in `ls /local/ssl/misc/*` > do > echo I found a config file $file > echo Its type is `/usr/bin/file $file` > done I found a config file /local/ssl/misc/CA.pl Its type is /local/ssl/misc/CA.pl: perl script text I found a config file /local/ssl/misc/CA.sh Its type is /local/ssl/misc/CA.sh: Bourne shell script text I found a config file /local/ssl/misc/c_hash Its type is /local/ssl/misc/c_hash: Bourne shell script text I found a config file /local/ssl/misc/c_info Its type is /local/ssl/misc/c_info: Bourne shell script text I found a config file /local/ssl/misc/c_issuer Its type is /local/ssl/misc/c_issuer: Bourne shell script text I found a config file /local/ssl/misc/c_name Its type is /local/ssl/misc/c_name: Bourne shell script text I found a config file /local/ssl/misc/der_chop Its type is /local/ssl/misc/der_chop: perl script text
This is how we insert the result of a shell command into a text string or variable.
cube$ loadTAB loadkeys loadmeter loadunimapThis shows the possible completions of commands which match "load". Type one more letter and TAB, and the rest will be filled in.
bash
csh
jsh
ksh
sh
sh5
tcsh
zsh
xterm
shelltool, cmdtool
screen
The best way to log onto another system is to use the Secure Shell
command ssh
. This replaces the now obsolete commands:
rlogin
rsh
telnet
These old commands are insecure andnote very flexible. The Secure Shell offers encryption, strong authentication and greater functionality. It can be used to run a single program on a remote machine, or to login on the remote machine.
cube$ ssh metaverse date cube$ ssh metaverse
ed
vi
ed
. This is the only "standard"
UNIX text editor supplied by vendors.
emacs
xemacs
pico
xedit
textedit
ls
dir
on other
systems).
cp
mv
touch
rm, unlink
mkdir, rmdir
cat
lp, lpr
lpq, lpstat
more
less
mc
kfm
chmod
chown, chgrp
chown
allows both these operations to be performed together using
the syntax chown owner.group file
.
acl
cut
paste
sed
awk
rmcr
find
locate
whereis
du
df
users
finger
who
w
write
talk
irc
mail
Mail
elm
pine
mailtool
rmail
netscape mail
zmail
tkrat
ftp
ncftp
cc
CC
gcc
g++
javac
java
ld
ar
dbx
gdb
xxgdb
ddd
perl
tcl
php
scheme
mercury
ps
vmstat
netstat
rpcinfo
showmount
uname
hostname
domainname
nslookup
archie, xarchie
xrn, fnews
netscape, xmosaic
tex, latex
texinfo
xdvi
dvips
ghostview, ghostscript
xv
xv -quit
to place a picture on your root window.
xpaint
xfig
xmgr
xsetroot
date
ispell
xcalc
dc,bc
xclock
ping
In order to communicate with a user, a shell needs to have access to a terminal. UNIX was designed to work with many different kinds of terminals. Input/output commands in UNIX read and write to a virtual terminal. In reality a terminal might be a text-based Teletype terminal (called a tty for short) or a graphics based terminal; it might be 80-characters wide or it might be wider or narrower. UNIX take into account these possibility by defining a number of instances of terminals in a more or less object oriented way.
Each user's terminal has to be configured before cursor based input/output will work correctly. Normally this is done by choosing one of a number of standard terminal types a list which is supplied by the system. In practice the user defines the value of the environment variable `TERM' to an appropriate name. Typical examples are `vt100' and `xterm'. If no standard setup is found, the terminal can always be configured manually using UNIX's most cryptic and opaque of commands: `stty'.
The job of configuring terminals is much easier now that hardware is more standard. Users' terminals are usually configured centrally by the system administrator and it is seldom indeed that one ever has to choose anything other than `vt100' or `xterm'.
Because UNIX originated before windowing technology was available, the user-interface was not designed with windowing in mind. The X window system attempts to be like a virtual machine park, running a different program in each window. Although the programs appear on one screen, they may in fact be running on UNIX systems anywhere in the world, with only the output being local to the user's display. The standard shell interface is available by running an X client application called `xterm' which is a graphical front-end to the standard UNIX textual interface.
The `xterm' program provides a virtual terminal using the X windows graphical user interface. It works in exactly the same way as a tty terminal, except that standard graphical facilities like copy and paste are available. Moreover, the user has the convenience of being able to run a different shell in every window. For example, using the `rlogin' command, it is possible to work on the local system in one window, and on another remote system in another window. The X-window environment allows one to cut and paste between windows, regardless of which host the shell runs on.
The X11 system is based on the client-server model. You might wonder why a window system would be based on a model which was introduced for interprocess communication, or network communication. The answer is straightforward.
The designers of the X window system realized that network communication was to be the paradigm of the next generation of computer systems. They wanted to design a system of windows which would enable a user to sit at a terminal in Massachusetts and work on a machine in Tokyo -- and still be able to get high quality windows displayed on their terminal. The aim of X windows from the beginning is to create a distributed window environment.
When I log onto my friend's Hewlett Packard workstation to use the text
editor (because I don't like the one on my EUNUCHS workstation) I want
it to work correctly on my screen, with my keyboard -- even though my
workstation is manufactured by a different company. I also want the
colours to be right despite the fact that the HP machine uses a
completely different video hardware to my machine. When I press the
curly brace key {
, I want to see a curly brace, and not some
hieroglyphic because the HP station uses a different keyboard.
These are the problems which X tries to address. In a network environment we need a common window system which will work on any kind of hardware, and hide the differences between different machines as far as possible. But it has to be flexible enough to allow us to change all of the things we don't like -- to choose our own colours, and the kind of window borders we want etc. Other windowing systems (like Microsoft windows) ignore these problems and thereby lock the user to a single vendors products and a single operating system. (That, of course, is no accident.)
The way X solves this problem is to use the client server model. Each program which wants to open a window on somebody's compute screen is a client of the X window service. To get something drawn on a user's screen, the client asks a server on the host of interest to draw windows for it. No client ever draws anything itself -- it asks the server to do it on its behalf. There are several reasons for this:
In X, the window manager is a different program to the server which does the drawing of graphics -- but the client-server idea still applies, it just has one more piece to its puzzle.
The X windows system is large and complex and not particularly user friendly. When you log in to the system, X reads two files in your home directory which decide which applications will be started what they will look like. The files are called
#!/bin/bash # # .xsession file # # PATH="/usr/bin:/bin:/local/gnu/bin:/usr/X11R6/bin" # # List applications here, with & at the end # so they run in the background # xterm -T NewTitle -sl 1000 -geometry 90x45+16+150 -sb & xclock & xbiff -geometry 80x80+510+0 & netscape -iconic& # Start a window manager. Exec replaces this script with # the fvwm process, so that it doesn't exist as a separate # (useless) process. exec /local/bin/fvwm
xterm*background: LightGrey Emacs*background: grey92 Xemacs*background: grey92
In the terminology used by X11, every client program has to contact a display in order to open a window. A display is a virtual screen which is created by the X server on a particular host. X can create several separate displays on a given host, though most machines only have one.
When an X client program wants to open a window, it looks in the UNIX environment variable `DISPLAY' for the IP address of a host which has an X server it can contact. For example, if we wrote
DISPLAY="myhost:0" export DISPLAY
the client would try to contact the X server on `myhost' and ask for a window on display number zero (the usual display). If we wrote
DISPLAY="198.112.208.35:0" export DISPLAY
the client would try to open display zero on the X server at the host with the IP address `198.112.208.35'.
Clearly there must be some kind of security mechanism to prevent just anybody from opening windows on someone's display. X has two such mechanisms:
xhost yourhost
would allow anyone using yourhost to
access the local display. This mechanism is only present for backward
compatibility with early versions of X windows. Normally one should
use the command xhost -
to exclude all others from accessing the
display.
xauth
is an
interactive utility used for controlling the contents of the `.Xauthority'
file. See the `xauth' manual page for more information.
The window paradigm has been very successful in many ways, but anyone who has used a window system knows that the screen is simply not big enough for all the windows one would like! UNIX has several solutions to this problem.
One solution is to attach several physical screens to a terminal. The X window system can support any number of physical screens of different types. A graphical designer might want a high resolution colour screen for drawing and a black and white screen for writing text, for instance. The disadvantage with this method is the cost of the hardware.
A cheaper solution is to use a window manager such as `fwvm' which creates a virtual screen of unlimited size on a single monitor. As the mouse pointer reaches the edge of the true screen, the window manager replaces the display with a new "blank screen" in which to place windows. A miniaturized image of the windows on a control panel acts as a map which makes it possible to find the applications on the virtual screen.
Yet another possibility is to create virtual displays inside a single window. In other words, one can collapse several shell windows into a single `xterm' window by running the program `screen'. The screen command allows you to start several shells in a single window (using CTRL-a CTRL-c) and to switch between them (by typing CTRL-a CTRL-n). It is only possible to see one shell window at a time, but it is still possible to cut and paste between windows and one has a considerable saving of space. The `screen' command also allows you to suspend a shell session, log out, log in again later and resume the session precisely where you left off.
Here is a summary of some useful screen commands:
To prevent all users from being able to access all files on the system, UNIX records information about who creates files and also who is allowed to access them later.
Each user has a unique username or loginname together with
a unique user id or uid. The user id is a number, whereas the login
name is a text string -- otherwise the two express the same information.
A file belongs to user A if it is owned by user A. User A then
decides whether or not other users can read, write or execute the
file by setting the protection bits or the permission of the
file using the command chmod
.
In addition to user identities, there are groups of users. The idea of
a group is that several named users might want to be able to read
and work on a file, without other users being able to access it.
Every user is a member of at least one group, called the login
group and each group has both a textual name and a number (group id).
The uid and gid of each user is recorded in the
file /etc/passwd
(See chapter 6). Membership of other groups
is recorded in the file /etc/group
or on some systems /etc/logingroup
.
The following output is from
the command ls -lag
executed on a SunOS type machine.
lrwxrwxrwx 1 root wheel 7 Jun 1 1993 bin -> usr/bin -r--r--r-- 1 root bin 103512 Jun 1 1993 boot drwxr-sr-x 2 bin staff 11264 May 11 17:00 dev drwxr-sr-x 10 bin staff 2560 Jul 8 02:06 etc drwxr-sr-x 8 root wheel 512 Jun 1 1993 export drwx------ 2 root daemon 512 Sep 26 1993 home -rwxr-xr-x 1 root wheel 249079 Jun 1 1993 kadb lrwxrwxrwx 1 root wheel 7 Jun 1 1993 lib -> usr/lib drwxr-xr-x 2 root wheel 8192 Jun 1 1993 lost+found drwxr-sr-x 2 bin staff 512 Jul 23 1992 mnt dr-xr-xr-x 1 root wheel 512 May 11 17:00 net drwxr-sr-x 2 root wheel 512 Jun 1 1993 pcfs drwxr-sr-x 2 bin staff 512 Jun 1 1993 sbin lrwxrwxrwx 1 root wheel 13 Jun 1 1993 sys->kvm/sys drwxrwxrwx 6 root wheel 732 Jul 8 19:23 tmp drwxr-xr-x 27 root wheel 1024 Jun 14 1993 usr drwxr-sr-x 10 bin staff 512 Jul 23 1992 var -rwxr-xr-x 1 root daemon 2182656 Jun 4 1993 vmUNIX
The first column is a textual representation of the protection bits for
each file. Column two is the number of hard links to the file (See exercises
below). The third and fourth columns are the user name and group name
and the remainder show the file size in bytes and the creation date.
Notice that the directories /bin
and /sys
are
symbolic links to other directories.
There are sixteen protection bits for a UNIX file, but only twelve of them can be changed by users. These twelve are split into four groups of three. Each three-bit number corresponds to one octal number.
The leading four invisible bits gives information about the type of file: is
the file a plain file, a directory or a link. In the
output from ls
this is represented by a single character:
-
, d
or l
.
The next three bits set the so-called s-bits and t-bit which are explained below.
The remaining three groups of three bits set flags which indicate whether a file can be read `r', written to `w' or executed `x' by (i) the user who created them, (ii) the other users who are in the group the file is marked with, and (iii) any user at all.
For example, the permission
Type Owner Group Anyone d rwx r-x ---
tells us that the file is a directory, which can be read and written to by the owner, can be read by others in its group, but not by anyone else.
Note about directories. It is impossible to cd
to a
directory unless the x
bit is set. That is, directories must be
`executable' in order to be accessible.
Here are some examples of the relationship between binary, octal and the textual representation of file modes.
Binary Octal Text 001 1 x 010 2 w 100 4 r 110 6 rw- 101 5 r-x - 644 rw-r--r--
It is well worth becoming familiar with the octal number representation of these permissions.
The chmod
command changes the permission or mode of a file. Only
the owner of the file or the superuser can change the permission.
Here are some examples of its use. Try them.
# make read/write-able for everyone chmod a+w myfile # add the 'execute' flag for directory chmod u+x mydir/ # open all files for everyone chmod 755 * # set the s-bit on my-dir's group chmod g+s mydir/ # descend recursively into directory opening all files chmod -R a+r dir
When a new file gets created, the operating system must decide what
default protection bits to set on that file. The variable umask
decides this.
umask
is normally set by each user in his or her .cshrc
file (see next chapter). For example
umask 077 # safe umask 022 # liberal
According the UNIX documentation, the value of umask
is
`XOR'ed (exclusive `OR') with a value of 666 & umask
for plain files or 777 & umask
for directories in order to find
out the standard protection. Actually this is not quite true: `umask'
only removes bits, it never sets bits which were not already set
in 666
. For instance
umask Permission 077 600 (plain) 077 700 (dir) 022 644 (plain) 022 755 (dir)
The correct rule for computing permissions is not XOR but `NOT AND'.
A UNIX program is normally executed by typing its pathname.
If the x
execute bit is not set on the file, this will generate
a `Permission denied' error. This protects the system from
interpreting nonsense files as programs. To make a program executable
for someone, you must therefore ensure that they can execute
the file, using a command like
chmod u+x filename
This command would set execute permissions for the owner of the file;
chmod ug+x filename
would set execute permissions for the owner and for any users in the same group as the file. Note that script programs must also be readable in order to be executable, since the shell has the interpret them by reading.
These two commands change the ownership and the group ownership of a file. Only the superuser can change the ownership of a file on most systems. This is to prevent users from being able to defeat quota mechanisms. (On some systems, which do not implement quotas, ordinary users can give a file away to another user but not get it back again.) The same applies to group ownership.
Normally users other than root cannot define their own groups. This is a weakness in UNIX from older times which no one seems to be in a hurry to change.
The s
and t
bits have special uses. They are described
as follows.
Octal Text Name 4000 chmod u+s Setuid bit 2000 chmod g+s Setgid bit 1000 chmod +t Sticky bit
The effect of these bits differs for plain files and directories and
differ between different versions of UNIX. You should check the manual
page man sticky
to find out about your system! The following is
common behaviour.
For executable files, the setuid bit tells UNIX that regardless of
who runs the program it should be executed with the permissions and
rights of owner of the file. This is often used to allow normal users
limited access to root
privileges. A setuid-root program
is executed as root
for any user. The setgid bit sets the group
execution rights of the program in a similar way.
In BSD UNIX, if the setgid bit is set on a directory then any new files created in that directory assume the group ownership of the parent directory and not the logingroup of the user who created the file. This is standard policy under system 5.
A directory for which the sticky bit is set restrict the deletion of
files within it. A file or directory
inside a directory with the t-bit set can
only be deleted or renamed by its owner or the superuser. This is
useful for directories like the mail spool area and /tmp
which must be writable to everyone, but should not allow a user
to delete another user's files.
(Ultrix) If an executable file is marked with a sticky bit, it is held in the
memory or system swap area. It does not have to be fetched from
disk each time it is executed. This saves time for frequently
used programs like ls
.
(Solaris 1) If a non-executable file is marked with the sticky bit, it will not be held in the disk page cache -- that is, it is never copied from the disk and held in RAM but is written to directly. This is used to prevent certain files from using up valuable memory.
On some systems (e.g. ULTRIX), only the superuser can set the sticky bit. On others (e.g. SunOS) any user can create a sticky directory.
The Bourne Again shell (Bash) is the command interpreter which you use to run programs and utilities. It contains a simple programming language for writing tailor-made commands, and allows you to join together UNIX commands with pipes. It is a configurable environment, and once you know it well, it is the most efficient way of working with UNIX.
The Bourne Again shell was written by the Free Software Foundation as a part of the GNU project and Bash is the default shell in most GNU/Linux distributions. Because of its command line editing features, it is much more efficient for interactive use than Bourne shell, the original UNIX shell. Most of the system scripts in UNIX are written in the Bourne shell. Although Bash includes many extensions and features not found in the Bourne shell, it maintains compatibility with it so that you can run Bourne shell scripts under Bash. On many GNU/Linux systems Bourne shell (`/bin/sh') is symbolically linked to Bash (`/bin/bash') so that the scripts that require the presence of the Bourne shell still run. If you want to write a platform independent shell script able to run on as many UNIX variants as possible, you should stick to Bourne shell syntax and avoid the Bash extensions.
When you log on to a GNU/Linux system and your login shell is defined in `/etc/passwd' to be Bash, it first executes commands in the `/etc/profile' file. It then searches for the `~/.bash_profile', `~/.bash_login' or `~/.profile' file, in this order, and executes commands in the first of these that is found and is readable. When a login exits, it executes commands in the `~/.bash_logout' file.
When you start an non-login interactive Bash shell, it only executes commands in the `~/.bashrc' file, if it exists and is readable. However, this shell inherits any environment (exported) variables from the parent shell, so environment variables set in `/etc/profile' and `~/.bash_profile' are passed onto the non-login shells and later to its subshells.
Here is a very simple example `~/.bashrc' file:
# # .bashrc - read in by every bash that starts. # umask 077 # Set the default file creation mask PATH="~/bin:$PATH" # Inserts own bin directory first in PATH PS1="`uname`:\h\$ " # prompt PS2="\h > " # prompt for foreach and while PRINTER=myprinter # Aliases are shortcuts to UNIX commands alias h=history alias ll="ls -l" alias cp='cp -i' alias rm='rm -i' alias c='ssh cube'
In order to make sure your `~/.bashrc' file is read when logging on
with ssh
to another machine, you may start your `~/.bash_profile'
file like this:
# # .bash_profile - read in every login. # if [ -f ~/.bashrc ] then source ~/.bashrc # runs .bashrc as if they where # typed into this file fi
Shell variables are defined using the syntax
VARIABLE="username is" myname="`whoami`"
It is important that there be no space between the variable and the equals sign. These variables are then referred to using the dollar `$' symbol.
$ echo "My $VARIABLE $myname" My username is mark
When assigning values to variables the dollar symbol is never used. By default these variables are local - that is they will not be passed on to programs and sub-shells running under the current shell. To make them global (so that child processes will inherit them) we use the command
export VARIABLE
This adds the variable to the process environment. Under Bash (but not under the old Bourne shell) it is also possible to declare a variable to be global on a single line by
export GLOBALVAR="global"
The command
set -a
changes the default so that all variables, after the command are created global.
Arrays or lists are often simulated in Bourne shell by sandwiching the colon `:' symbol between items
PATH=/bin:/usr/bin:/etc:/local/bin:. LD_LIBARAY_PATH=/usr/lib:/usr/openwin/lib:/local/lib
but there is no real facility for arrays in the Bourne shell. Note that the UNIX `cut' command can be used to extract the elements of the list. Loops can also read such lists directly See section Loops in Bash. However, Bash version 2.x supports arrays as seen in the next section.
The value of a variable is given by the dollar symbol. It is also possible to use curly braces around the variable name to `protect' the variable from interfering text. For example:
$ animal=worm $ echo book$animal bookworm $ thing=book $ echo $thingworm (nothing..) $ echo ${thing}worm bookworm
Default values can be given to variables in the Bourne shell. The following commands illustrate this.
echo ${var-"No value set"} echo ${var="Octopus"} echo ${var+"Forced value"} echo ${var?"No such variable"}
The first of these prints out the contents of `$var', if it is defined. If it is not defined the variable is substituted for the string "No value set". The value of `var' is not changed by this operation. It is only for convenience.
The second command has the same effect as the first, but here the value of `$var' is actually changed to "Octopus" if `$var' is not set.
The third version is slightly peculiar. If `$var' is already set, its value will be forced to be "Forced value", otherwise it is left undefined.
Finally the last instance issues an error message "No such variable" if `$var' is not defined.
In Bash 2.x it is possible to extract parts of the string a variable
is set to using the
construction ${variable:offset:length}|
as shown in the next example.
var="abcdefg" middle=${var:2:3} echo $middle cde
An offset of 2 skips the first 2 characters and a string of length 3 is extracted from the middle of the string.
The original Bourne shell does not have arrays. Bash version 2.x does have arrays, however. An array can be assigned from a string of words separated by whitespaces or the individual elements of the array can be set individually.
colours=(red white green) colours[3]="yellow"
An element of the array must be referred to using curly braces.
echo ${colours[1]} white
Note that the first element of the array has index 0. The set of
all elements is referred to by ${colours[*]}
.
echo ${colours[*]} red white green yellow echo ${#colours[*]} 4
As seen the number of elements in an array is given by ${#colours[*]}
.
When the shell starts up, it inherits three files: `stdin', `stdout', and `stderr'. Standard input normally comes from the keyboard. Standard output and standard error normally go to the screen. There are times you want to read input from a file or send output of errors to a file. This can be accomplished by using I/O redirection.
In Bash and the Bourne shell, the standard input/output files are referred to by numbers rather than by names.
The default routes for these files can be changed by redirection. The output of the
command echo
is by default sent to the screen, that is the stdout with
file number 1 is sent to the screen. Using redirection operators it is possible
to redirect the standard out of echo
to where we want it. We can send
output to a file with the following command.
echo "should be sent to a file" > file.txt
This creates a new file `file.txt' containing the string 'should be sent to a file'.
The redirection operator could have been given as 1>
, but it is understood
that standard out is meant when skipping the number of the file handle.
The single '>' always creates a new file, while '>>' appends to the end of a file.
If you had mistyped the command echo
the result would have been:
ehco "should be sent to a file" > file.txt bash: ehco: command not found
The standard error with file handle 2 is by default sent to the screen, independent of where standard out (1) is sent. If you like you can redirect stdout to another or the same file.
ehco "should be sent to a file" > file.txt 2> error.txt cat error.txt bash: ehco: command not found
There are several ways to send stderr to the same file as stdin is redirected to. The following three commands are equivalent.
ehco "should be sent to a file" >& file.txt ehco "should be sent to a file" > file.txt 2> file.txt ehco "should be sent to a file" > file.txt 2>&1
The string 2>&1
means that stderr(2) should be sent to the same file as
stdout(1). This is the only why to do this under the Bourne shell and this construction
is therefore often seen in system shell scripts.
Furthermore it is possible to force a command which by default takes standard input from the keyboard, to read input from a file by redirecting stdin. The mail-command expects input from keyboard, but the '<' redirection operator makes it send the password file to the user mark:
/bin/mail mark < /etc/passwd
The following table summarizes the most important redirection operators:
Redirection operator What it does < Redirects input > Redirects output >> Appends output 2> Redirects error >& Redirects output and error (Bash only) 2>&1 Redirects error where output (1) is going
A pipe takes the output from the command on the left-hand side of the
pipe symbol and sends it to the input of the command on the right-hand side
of the pipe symbol. A pipeline can consist of several pipes and this makes
pipes a very powerful tool. It enables us to combine all the small and
efficient UNIX commands in any thinkable way. If you want to count the number
of people logged on, you could save the output of the command who
in the temporary file `tmp', use wc -l
to count the number
of lines in `tmp' and finally remove the temporary file.
$ who > tmp $ wc -l tmp 4 tmp $ rm tmp
Using a pipe saves disk space and time: the stdout from who
can be
redirected to the stdin of wc -l
through a pipe and there is no need
for temporarily storing the output from who
.
$ who | wc -l 4
Most UNIX-commands are constructed with piping in mind and this makes it possible to solve complex tasks easily, by joining commands along a pipeline. Consider the following pipeline:
cat big.jpg | djpeg | pnmscale -pixels 150000 | cjpeg > small.jpg
The command cat
sends the large JPEG-image to
djpeg
which decompresses it and sends the resulting bitmap to
stdout. The stream of data floats through the next pipe to pnmscale
which scales the bitmap image down to the given size. The scaled image is piped to
the command cjpeg
which compresses the standard input and finally
produces a JPEG-image of reduced size which is stored in the file
`small.jpg'.
The history feature in Bash means that you do not have to type commands over and over again. You can use the UP ARROW key to browse back through the list of commands you have typed previously and the keys LEFT ARROW and RIGHT ARROW to edit these commands.
In addition there are a couple of commands which selects commands from the history list.
The first of these simply repeats the last command. The second command gives an absolute number. The absolute command number can be seen by typing `history'.
In Bash you can save hours worth of typing errors by using the completion mechanism. This feature is based on the TAB key.
The idea is that if you type half a filename and press TAB, the shell will try to guess the remainder of the filename. It does this by looking at the files which match what you have already typed and trying to fill in the rest. If there are several files which match, the shell sounds the "bell" or beeps. You can then type TAB twice to obtain a list of the possible alternatives. Here is an example: suppose you have just a single file in the current directory called `very_long_filename', typing
more TAB
results in the following appearing on the command line
more very_long_filename
The shell was able to identify a unique file. Now suppose that you have two files called `very_long_filename' and `very_big_filename', typing
more TAB
results in the following appearing on the command line
more very_
and the shell beeps, indicating that the choice was not unique and a decision is required. Next, you type TAB twice(3) to see which files you have to choose from and the shell lists them and returns you to the command line, exactly where you were. You now choose `very_long_filename' by typing `l'. This is enough to uniquely identify the file. Pressing the TAB key again results in
more very_long_filename
on the screen. As long as you have written enough to select a file uniquely, the shell will be able to complete the name for you.
Completion also works on shell commands, but it is a little slower since the shell must search through all the directories in the command path to complete commands.
Two kinds of quotes can be used in shell apart from the backward quotes we mentioned above. The essential difference between them is that certain shell commands work inside double quotes but not inside single quotes. For example
cube$ echo /etc/rc* /etc/rc.boot /etc/rc0.d /etc/rc1.d /etc/rc2.d /etc/rc3.d /etc/rc4.d cube$ echo "/etc/rc*" /etc/rc* cube$ echo "`whoami` -- my name is $USER" mark -- my name is mark cube$ echo '`whoami` -- my name is $USER' `whoami` -- my name is $USER
We see that the single quotes prevent variable substitution and sub-shells. Wildcards do not work inside either single or double quotes.
So far we haven't mentioned UNIX's ability to multitask. In the Bourne shell (`sh') there are no facilities for controlling several user processes. Bash provides some commands for starting and stopping processes. These originate from the days before windows and X11, so some of them may seem a little old-fashioned. They are still very useful nonetheless.
Let's begin by looking at the commands which are true for any shell. Most programs are run in the foreground or interactively. That means that they are connected to the standard input and send their output to the standard output. A program can be made to run in the background, if it does not need to use the standard I/O. For example, a program which generates output and sends it to a file could run in the background. In a window environment, programs which create their own windows can also be started as background processes, leaving standard I/O in the shell free.
Background processes run independently of what you are doing in the foreground.
A background process is started using the special character `&' at the end of the command line.
find / -name '*lib*' -print >& output &
The final `&' on the end of this line means that the job will be run in the background. Note that this is not confused with the redirection operator `>&' since it must be the last character on the line. The command above looks for any files in the system containing the string `lib' and writes the list of files to a file called `output'. This might be a useful way of searching for missing libraries which you want to include in your environment variable `LD_LIBRARY_PATH'. Searching the entire disk from the root directory `/' could take a long time, so it pays to run this in the background.
If we want to see what processes are running, we can use the `ps' command. `ps' without any arguments lists all of your processes, i.e. all processes owned by the user name you logged in with in the current shell. `ps' takes many options, for instance `ps auxg' will list all processes in gruesome detail (The "g" is for group, not gruesome!). `ps' reads the kernel's process tables directly.
Processes can be stopped and started, or killed one and for all. The `kill' command does this. There are, in fact, two versions of the `kill' command. One of them is built into Bash and the other is not. If you use Bash then you will never care about the difference. We shall nonetheless mention the special features of Bash built-ins below. The kill command takes a number called a signal as an argument and another number called the process identifier or PID for short. Kill send signals to processes. Some of these are fatal and some are for information only. The two commands
kill -15 127 kill 127
are identical. They both send signal 15 to PID 127. This is the normal termination signal and it is often enough to stop any process from running.
Programs can choose to ignore certain signals by trapping signals with a special handler. One signal they cannot ignore is signal 9.
kill -9 127
is a sure way of killing PID 127. Even though the process dies, it may not be removed from the kernel's process table if it has a parent (see next section).
Here is the complete list of signals which the Linux kernel send to processes in different circumstances.
#define SIGHUP 1 /* Hangup (POSIX). */ #define SIGINT 2 /* Interrupt (ANSI). */ #define SIGQUIT 3 /* Quit (POSIX). */ #define SIGILL 4 /* Illegal instruction (ANSI). */ #define SIGTRAP 5 /* Trace trap (POSIX). */ #define SIGABRT 6 /* Abort (ANSI). */ #define SIGIOT 6 /* IOT trap (4.2 BSD). */ #define SIGBUS 7 /* BUS error (4.2 BSD). */ #define SIGFPE 8 /* Floating-point exception (ANSI). */ #define SIGKILL 9 /* Kill, unblockable (POSIX). */ #define SIGUSR1 10 /* User-defined signal 1 (POSIX). */ #define SIGSEGV 11 /* Segmentation violation (ANSI). */ #define SIGUSR2 12 /* User-defined signal 2 (POSIX). */ #define SIGPIPE 13 /* Broken pipe (POSIX). */ #define SIGALRM 14 /* Alarm clock (POSIX). */ #define SIGTERM 15 /* Termination (ANSI). */ #define SIGSTKFLT 16 /* Stack fault. */ #define SIGCLD SIGCHLD /* Same as SIGCHLD (System V). */ #define SIGCHLD 17 /* Child status has changed (POSIX). */ #define SIGCONT 18 /* Continue (POSIX). */ #define SIGSTOP 19 /* Stop, unblockable (POSIX). */ #define SIGTSTP 20 /* Keyboard stop (POSIX). */ #define SIGTTIN 21 /* Background read from tty (POSIX). */ #define SIGTTOU 22 /* Background write to tty (POSIX). */ #define SIGURG 23 /* Urgent condition on socket (4.2 BSD). */ #define SIGXCPU 24 /* CPU limit exceeded (4.2 BSD). */ #define SIGXFSZ 25 /* File size limit exceeded (4.2 BSD). */ #define SIGVTALRM 26 /* Virtual alarm clock (4.2 BSD). */ #define SIGPROF 27 /* Profiling alarm clock (4.2 BSD). */ #define SIGWINCH 28 /* Window size change (4.3 BSD, Sun). */ #define SIGPOLL SIGIO /* Pollable event occurred (System V). */ #define SIGIO 29 /* I/O now possible (4.2 BSD). */ #define SIGPWR 30 /* Power failure restart (System V). */ #define SIGSYS 31 /* Bad system call. */
We have already mentioned 15 and 9 which are the main signals for users. Signal 1, or `HUP' can be sent to certain programs by the superuser. For instance
kill -1 <inetd> kill -HUP <inetd>
which forces `inetd' to reread its configuration file. Sometimes it is useful to suspend a process temporarily and then restart it later.
kill -20 <PID> # suspend process <PID> kill -18 <PID> # resume process <PID>
When you start a process from a shell, regardless of whether it is a background process or a foreground process, the new process becomes a child of the original shell. Remember that the shell is just a UNIX process itself. Moreover, if one of the children starts a new process then it will be a child of the child (a grandchild?)! Processes therefore form hierarchies. Several children can have a common parent.
If we kill a parent, then (unless the child has detached itself from the parent) all of its children die too. If a child dies, the parent is not affected. Sometimes when a child is killed, it does not die but becomes "defunct" or a zombie process. This means that the child has a parent which is waiting for it to finish. If the parent has not yet been informed that the child has died, for example because it has been suspended itself, then the dead child is not removed from the kernel's process table. When the parent wakes up and receives the message that the child has terminated, the process entry for the dead child can be removed.
Now let's look at some commands which are built into Bash for starting and stopping processes. Bash refers to user programs as `jobs' rather than processes -- but there is no real difference. The added bonus of Bash is that each shell has a job number in addition to its PID. The job numbers are simpler and are private for the shell, whereas the PIDs are assigned by the kernel and are often very large numbers which are difficult to to remember. When a command is executed in the shell, it is assigned a job number. If you never run any background jobs then there is only ever one job number: 1, since every job exits before the next one starts. However, if you run background tasks, then you can have several jobs "active" at any time. Moreover, by suspending jobs, Bash allows you to have several interactive programs running on the same terminal -- the `fg' and `bg' commands allow you to move commands from the background to the foreground and vice-versa.
Take a look at the following shell session.
cube$ emacs myfile& [3] 771 cube$ ( other commands ... , edit myfile and close emacs )
When a background job is done, the shell prints a message at a suitable moment between prompts.
[3]+ Done emacs myfile cube$
This tells you that job number 1 finished normally. If the job exits abnormally then the word `Done' may be replaced by some other message. For instance, if you kill the job, it will say
cube$ kill %3 cube$ [3]+ Terminated emacs myfile cube$
You can list the jobs you have running using the `jobs' command. The output looks something like
cube$ jobs [1] Terminated xdvi unix [2] Running xemacs unix.texinfo & [3] Running xterm -sb -sl 10000 & [4] Running ghostview & [5] Running netscape & [6] Running xterm -sb -sl 10000 & [7] Running xemacs fil & [8]+ Stopped emacs unix.log [9]- Running gimp &
To suspend a program which you are running in the foreground you can type CTRL-z (this is like sending a `kill -20' signal from the keyboard). (4) You can suspend any number of programs and then restart them one at a time using `fg' and `bg'. If you want job 5 to be restarted in the foreground, you would type
fg %5
When you have had enough of job 5, you can type CTRL-z to suspend it and then type
fg %6
to activate job 6. Provided a job does not want to send output to `stdout', you can restart any job in the background, using a command like.
bg %4
This method of working was useful before windows were available. Using `fg' and `bg', you can edit several files or work on several programs without have to quit to move from one to another.
See also some related commands for batch processing `at', `batch' and `atq', `cron'.
NOTE: CTRL-c sends a `kill -2' signal, which send a standard interrupt message to a program. This is always a safe way to interrupt a shell command.
In Bourne shell arithmetic is performed entirely `by proxy'. To evaluate an expression we call the `expr' command or the `bc' precision calculator. Here are some examples of `expr'
a=`expr $a+1` # increment a a=`expr 4 + 10 \* 5` # 4+10*5 check = `expr $a \> $b` # true=1, false=0. True if $a > $b
`expr' is very sensitive to spaces and backslash characters and this makes it a bit awkward to do arithmetic under the Bourne shell.
Bash 2.0
provides a new and simpler way to do arithmetic using double parentheses.
If you surround any integer arithmetic expression as in
(( x = y + 1 ))
, you can perform most arithmetic operations with
the same syntax as in Java and C.
(( x = 1 )) echo $x 1 (( x++ )) (( y = 4*x )) echo $y 8
Note that you do not need to use the dollar symbol to refer to a variable within the double parentheses (but you may do it) and that spaces are allowed.
(( sum = 2 )) (( total = 4*$sum + sum )) echo $total 10
The variables within double parentheses are throughout treated as integers. Assigning a float value like 2.5 to a variable results in an syntax error while assigning a string to a variable cause the string to be stored as zero.
Scripts are created by making an executable file which begins with the sequence of characters
#!/bin/bash
This construction is quite general: any executable file which begins with a sequence
#!myprogram -option
will cause the shell to attempt to execute
myprogam -option filename
where filename is the name of the file.
If a script is to accept arguments then these can be referred to as ` $1 $2 $3..$9'. There is a logical limit of nine arguments to a Bourne script, but Bash handles the next arguments as `${10}'. `$0' is the name of the script itself.
Here is a simple Bash script which prints out all its arguments.
#!/bin/bash # # Print all arguments (version 1) # for arg in $* do echo Argument $arg done echo Total number of arguments was $#
The `$*' symbol stands for the entire list of arguments and `$#' is the total number of arguments.
Another way of achieving the same is to use the `shift' command. We shall meet this again in the Perl programming language. `shift' takes the first argument from the argument list and deletes it, moving all of the other arguments down one number -- this is how we can handle long lists of arguments in the Bourne shell.
#!/bin/bash # # Print all arguments (version 2) # while ( true ) do arg=$1; shift; echo $arg was an argument; if [ $# -eq 0 ]; then break fi done
All programs which execute in UNIX return a value through the C `return' command. There is a convention that a return value of zero (0) means that everything went well, whereas any other value implies that some error occurred. The return value is usually the value returned in `errno', the external error variable in C.
Shell scripts can test for these values either by placing the command directly inside an `if' test, or by testing the variable `$?' which is always set to the return code of the last command. Some examples are given following the next two sections.
Bash and the Bourne shell has an array of tests. They are written as follows. Notice that `test' is itself not a part of the shell, but is a program which works out conditions and provides a return code. See the manual page on `test' for more details.
test -f file
test -d file
test -r file
test -w file
test -x file
test -s file
test -g file
test -u file
test s1 = s2
test s1 != s2
test x -eq y
test x -ne y
test x -gt y
test x -lt y
test x -ge y
test x -le y
!
-a
-o
Note that an alternate syntax for writing these commands if to use the square brackets, instead of writing the word test.
[ $x -lt $y ] "==" test $x -lt $y
Just as with the arithmetic expressions, Bash 2.x provides a syntax for conditionals which are more similar to Java and C. While arithmetic C-like expressions can be used within double parentheses, C-like tests can be used within double square brackets.
[[ $var == "OK" || $var == "yes" ]]
This C-like syntax is not allowed in the Bourne shell, but is equivalent to
[ $var = "OK" -o $var = "yes" ]
which is valid in both shells.
Arithmetic C-like tests can be used within double parentheses so that under Bash 2.x the following tests are equivalent:
[ $x -lt $y ] "==" (( x < y ))
The conditional structures have the following syntax.
if UNIX-command then command else commands fi
The `else' clause is, of course, optional. As noted before, the first UNIX command could be anything, since every command has a return code. The result is TRUE if it evaluates to zero and false otherwise (in contrast to the conventions in most languages). Multiple tests can be made using
if UNIX-command then commands elif UNIX-command then commands elif UNIX-command then commands else commands fi
where `elif' means `else-if'.
The equivalent of the C-school's `switch' statement is a more Pascal-like `case' structure.
case UNIX-command-or-variable in wildcard1) commands ;; wildcard2) commands ;; wildcard3) commands ;; esac
This structure uses the wildcards to match the output of the command or variable in the first line. The first pattern which matches gets executed.
In shell you can read the value of a variable using the `read' command, with syntax
read variable
This reads in a string from the keyboard and terminates on a newline character. Under the old Bourne shell another way to do this is to use the `input' command to access a particular logical device. The keyboard device in the current terminal is `/dev/tty', so that one writes
variable = `line < /dev/tty`
which fetches a single line from the user. The command line
is however not available in most GNU/Linux distributions.
Here are some examples of these commands. First a program which asks yes or no...
#!/bin/bash # # Yes or no # echo "Please answer yes or no: " read answer case $answer in y* | Y* | j* | J* ) echo YES!! ;; n* | N* ) echo NO!! ;; *) echo "Can't you answer a simple question?" esac echo The end
Notice the use of pattern matching and the `|' `OR' symbol.
#!/bin/bash # # Kernel check # if test ! -f /vmUNIX # Check that the kernel is there! then echo "This is not BSD UNIX...hmmm" if [ -f /hp-ux ] then echo "It's a Hewlett Packard machine!" fi elif [ -w /vmUNIX ] then echo "HEY!! The kernel is writable my me!"; else echo "The kernel is write protected." echo "The system is safe from me today." fi
The loop structures in Bash and in the Bourne shell have the following syntax.
while UNIX-command do commands done
The first command will most likely be a test but, as before, it could in principle be any UNIX command. The `until' loop, reminiscent of BCPL, carries out a task until its argument evaluates to TRUE.
until UNIX-command do commands done
Finally the `for' structure has already been used above.
for variable in list do commands done
Often we want to be able to use an array of values as the list which
for
parses, but Bourne shell has no array variables. This
problem is usually solved by making a long string separated by, for
example, colons. For example, the $PATH
variable has the
form
PATH = /usr/bin:/bin:/local/gnu/bin
Bourne shell allows us to split such a string on whatever character we wish. Normally the split is made on spaces, but the variable `IFS' can be defined with a replacement. To make a loop over all directories in the command path we would therefore write
IFS=: for name in $PATH; do commands done
The best way to gain experience with these commands is through some examples.
#!/bin/bash # # Get text from user repeatedly # echo "Type away..." while read TEXT do echo You typed $TEXT if [ "$TEXT" = "quit" ]; then echo "(So I quit!)" exit 0 fi done echo "HELP!"
This very simple script is a typical use for a while-loop. It gets text repeatedly until the user type `quit'. Since read never returns `false' unless an error occurs or it detects an EOF (end of file) character CTRL-D, it will never exit without some help from an `if' test. If it does receive a CTRL-D signal, the script prints `HELP!'.
#!/bin/bash # # Watch in the background for a particular user # and give alarm if he/she logs in # # To be run in the background, using & # if [ $# -ne 1 ]; then echo "Give the name of the user as an argument" > /dev/tty exit 1 fi echo "Looking for $1" until users | grep -s $1 do sleep 60 done echo "!!! WAKE UP !!!" > /dev/tty echo "User $1 just logged in" > /dev/tty
This script uses `grep' in `silent mode' (-s option). i.e. grep never writes anything to the terminal. The only thing we are interested in is the return code the piped command produces. If `grep' detects a line containing the username we are interested in, then the result evaluates to TRUE and the sleep-loop exits.
Our final example is the kind of script which is useful for a system administrator. It transfers over the Network Information Service database files so that a slave server is up to date. All we have to do is make a list of the files and place it in a `for' loop. The names used below are the actual names of the NIS maps, well known to system administrators.
#!/bin/bash # # Update the NIS database maps on a client server. This program # shouldn't have to be run, but sometimes things go wrong and we # have to force a download from the main sever. # PATH=/etc/yp:/usr/etc/yp:$PATH MASTER=myNISserver for map in auto.direct auto.master ethers.byaddr ethers.byname\ group.bygid group.byname hosts.byaddr hosts.byname\ mail.aliases netgroup.byhost netgroup.byuser netgroup\ netid.byname networks.byaddr networks.byname passwd.byname\ passwd.byuid priss.byname protocols.byname protocols.bynumber\ rpc.bynumber services.byname services usenetgroups.byname; do ypxfr $1 -h $MASTER $map done
One of the worthy features of the Bourne shell is that it allows you to define subroutines or procedures. Subroutines work just like subroutines in any other programming language. They are executed in same shell (not as a sub-process).
Here is an interesting program which demonstrates two useful things at the same time. First of all, it shows how to make a hierarchical subroutine structure using the Bourne shell. Secondly, it shows how the `trap' directive can be used to trap signals, so that Bourne shell programs can exit safely when they are killed or when CTRL-C is typed.
#!/bin/bash # # How to make a signal handler in Bourne Shell # using subroutines # ##################################################### # Level 2 ##################################################### ReallyQuit() { while true do echo "Do you really want to quit?" read answer case $answer in y* | Y* ) return 0;; *) echo "Resuming..." return 1;; esac done } ##################################################### # Level 1 ##################################################### SignalHandler() { if ReallyQuit # Call a function then exit 0 else return 0 fi } ##################################################### # Level 0 : main program ##################################################### trap SignalHandler 2 15 # Trap kill signals 2 and 15 echo "Type some lines of text..." while read text do echo "$text - CTRL-C to exit" done
Note that the logical tree structure of this program is upside down (the highest level comes at the bottom). This is because all subroutines must be defined before they are used.
This example concludes our survey of Bash and the Bourne shell.
The superuser `root' is the only privileged user in UNIX. All other users have only restricted access to the system. Usually this is desirable, but sometimes it is a nuisance.
A setuid script is a script which has its setuid-bit set. When such a script is executed by a user, it is run with all the rights and privileges of the owner of the script. All of the commands in the script are executed as the owner of the file and not with the user-id of the person who ran the script. If the owner of the setuid script is `root' then the commands in the script are run with root privileges!
Setuid scripts are clearly a touchy security issue. When giving away one's rights to another user (especially those of `root') one is tempting hackers. Setuid scripts should be avoided.
A setgid program is almost the same, but only the group id is set to that of the owner of the file. Often the effect is the same.
An example of a setuid program is the `ps' program. `ps' lists all of the processes running in the kernel. In order to do this it needs permission to access the private data structures in the kernel. By making `ps' setgid root, it allows ordinary users to be able to read as much as the writers of `ps' thought fit, but no more.
Naturally, only the superuser can make a file setuid or setgid root.
Programmers who are used to C or C++ often find it easier to program in C-shell because there are strong similarities between the two.
Most users run the C-shell `/bin/csh' as their login environment,
or these days, preferably the `tcsh' which is an improved version
of csh. When a user logs in to a UNIX system the C-shell starts by
reading some files which configure the environment by defining
variables like path
.
With the advent of the X11 windowing system, this has changed slightly. Since the window system takes over the entire login procedure, users never get to run `login shells', since the login shell is used up by the X11 system. On an X-terminal or host running X the `.login' file normally has no effect.
With some thought, the `.login' file can be eliminated entirely,
and we can put everything into the .cshrc
file.
Here is a very simple example `.cshrc' file.
# # .cshrc - read in by every csh that starts. # # Set the default file creation mask umask 077 # Set the path set path=( /usr/local/bin /usr/bin/X11 /usr/ucb /bin /usr/bin . ) # Exit here if the shell is not interactive if ( $?prompt == 0 ) exit # Set some variables set noclobber notify filec nobeep set history=100 set prompt="`hostname`%" set prompt2 = "%m %h>" # tcsh, prompt for foreach and while setenv PRINTER myprinter setenv LD_LIBRARY_PATH /usr/lib:/usr/local/lib:/usr/openwin/lib # Aliases are shortcuts to UNIX commands alias passwd yppasswd alias dir 'ls -lg \!* | more' alias sys 'ps aux | more' alias h history
It is possible to make a much more complicated .cshrc file than this. The advent of distributed computing and NFS (Network file system) means that you might log into many different machines running different versions of UNIX. The command path would have to be set differently for each type of machine.
We have already seen in the examples above how to define variables in C-shell. Let's formalize this. To define a local variable -- that is, one which will not get passed on to programs and sub-shells running under the current shell, we write
set local = "some string" set myname = "`whoami`"
These variables are then referred to by using the dollar `$' symbol. i.e. The value of the variable `local' is `$local'.
echo $local $myname
Global variables, that is variables which all sub-shells inherit from the current shell are defined using `setenv'
setenv GLOBAL "Some other string" setenv MYNAME "`who am i`"
Their values are also referred to using the `$' symbol. Notice that
set
uses an `=' sign while `setenv' does not.
Variables can be also created without a value. The shell uses this method to switch on and off certain features, using variables like `noclobber' and `noglob'. For instance
nexus% set flag nexus% if ($?flag) echo 'Flag is set!' Flag is set! nexus% unset flag nexus% if ( $?flag ) echo 'Flag is set!' nexus%
The operator `$?variable' is `true' if variable exists and `false' if it does not. It does not matter whether the variable holds any information.
The commands `unset' and `unsetenv' can be used to undefine or delete variables when you don't want them anymore.
A useful facility in the C-shell is the ability to make arrays out of strings and other variables. The round parentheses `(..)' do this. For example, look at the following commands.
nexus% set array = ( a b c d ) nexus% echo $array[1] a nexus% echo $array[2] b nexus% echo $array[$#array] d nexus% set noarray = ( "a b c d" ) nexus% echo $noarray[1] a b c d nexus% echo $noarray[$#noarray] a b c d
The first command defines an array containing the elements `a b c d'. The elements of the array are referred to using square brackets `[..]' and the first element is `$array[1]'. The last element is `$array[4]'. NOTE: this is not the same as in C or C++ where the first element of the array is the zeroth element!
The special operator `$#' returns the number of elements in an array. This gives us a simple way of finding the end of the array. For example
nexus% echo $#path 23 nexus% echo "The last element in path is $path[$#path]" The last element in path is .
To find the next last element we need to be able to do arithmetic. We'll come back to this later.
The symbols
< > >> << | &
have a special meaning in the shell. By default, most commands take their input from the file `stdin' (the keyboard) and write their output to the file `stdout' and their error messages to the file `stderr' (normally, both of these output files are defined to be the current terminal device `/dev/tty', or `/dev/console').
`stdin', `stdout' and `stderr', known collectively as `stdio', can be redefined or redirected so that information is taken from or sent to a different file. The output direction can be changed with the symbol `>'. For example,
echo testing > myfile
produces a file called `myfile' which contains the string `testing'. The single `>' (greater than) sign always creates a new file, whereas the double `>>' appends to the end of a file, if it already exists. So the first of the commands
echo blah blah >> myfile echo Newfile > myfile
adds a second line to `myfile' after `testing', whereas the second command writes over `myfile' and ends up with just one line `Newfile'.
Now suppose we mistype a command
ehco test > myfile
The command `ehco' does not exist and so the error message `ehco: Command not found' appears on the terminal. This error message was sent to stderr -- so even though we redirected output to a file, the error message appeared on the screen to tell us that an error occurred. Even this can be changed. `stderr' can also be redirected by adding an ampersand `&' character to the `>' symbol. The command
ehco test >& myfile
results in the file `myfile' being created, containing the error message `ehco: Command not found'.
The input direction can be changed using the `<' symbol for example
/bin/mail mark < message
would send the file `message' to the user `mark' by electronic mail. The mail program takes its input from the file instead of waiting for keyboard input.
There are some refinements to the redirection symbols. First of all, let us introduce the C-shell variable `noclobber'. If this variable is set with a command like
set noclobber
then files will not be overwritten by the `>' command. If one tries to redirect output to an existing file, the following happens.
UNIX% set noclobber UNIX% touch blah # create an empty file blah UNIX% echo test > blah blah: File exists.
If you are nervous about overwriting files, then you can set `noclobber' in your `.cshrc' file. `noclobber' can be overridden using the pling `!' symbol. So
UNIX% set noclobber UNIX% touch blah # create an empty file blah UNIX% echo test >! blah
writes over the file `blah' even though `noclobber' is set.
Here are some other combinations of redirection symbols
The last of these commands reads from the standard input until it finds a line which contains a word. It then feeds all of this input into the program concerned. For example,
nexus% mail mark <<quit nexus 1> Hello mark nexus 2> Nothing much to say... nexus 2> so bye nexus 2> nexus 2> quit Sending mail... Mail sent!
The mail message contains all the lines up to, but not including `marker'. This method can also be used to print text verbatim from a file without using multiple echo commands. Inside a script one may write:
cat << "marker"; MENU 1) choice 1 2) choice 2 ... marker
The cat
command writes directly to stdout
and the
input is redirected and taken directly from the script file.
A very useful construction is the `pipe' facility. Using the `|' symbol one can feed the `stdout' of one program straight into the `stdin' of another program. Similarly with `|&' both `stdout' and `stderr' can be piped into the input of another program. This is very convenient. For instance, look up the following commands in the manual and try them.
ps aux | more echo 'Keep on sharpening them there knives!' | mail henry vmstat 1 | head ls -l /etc | tail
Note that when piping both standard input and standard error to another program, the two files do not mix synchronously. Often `stderr' appears first.
Occasionally you might want to have a copy of what you see on your terminal sent to a file. `tee' and `script' do this. For instance,
find / -type l -print | tee myfile
sends a copy of the output of `find' to the file `myfile'. `tee' can split the output into as many files as you want:
command | tee file1 file2 ....
You can also choose to record the output an entire shell session using the `script' command.
nexus% script mysession Script started, file is mysession nexus% echo Big brother is scripting you Big brother is scripting you nexus% exit exit Script done, file is mysession
The file `mysession' is a text file which contains a transcript of the session.
One of the useful features of the shell is that you can use the normal UNIX commands to make programs called scripts. To make a script, you just create a file containing shell commands you want to execute and make sure that the first line of the file looks like the following example.
#!/bin/csh -f # # A simple script: check for user's mail # # set path = ( /bin /usr/ucb ) # Set the local path cd /var/spool/mail # Change dir foreach uid ( * ) echo "$uid has mail in the intray! " # space prevents an error! end
The sequence `#!/bin/csh' means that the following commands are to be fed into `/bin/csh'. The two symbols `#!' must be the very first two characters in the file. The `-f' option means that your `.cshrc' file is not read by the shell when it starts up. The file containing this script must be executable (see `chmod') and must be in the current path, like all other programs.
Like C programs, C-shell scripts can accept command line arguments. Suppose you want to make a program to say hello to some other users who are logged onto the system.
say-hello mark sarah mel
To do this you need to know the names that were typed on the command line. These names are copied into an array in the C-shell called the argument vector, or `argv'. To read these arguments, you just treat `argv' as an array.
#!/bin/csh -f # # Say hello # foreach name ( $argv ) echo Saying hello to $name echo "Hello from $user! " | write $name end
The elements of the array can be referred to as `argv[1]'..`argv[$#argv]' as usual. They can also be referred to as `$1'..`$3' up to the last acceptable number. This makes C-shell compatible with the Bourne shell as far as arguments are concerned. One extra flourish in this method is that you can also refer to the name of the program itself as `$0'. For example,
#!/bin/csh -f echo This is program $0 running for $user
`$argv' represents all the arguments. You can also use `$*' from the Bourne shell.
The C-shell does not allow you to define subroutines or functions, but you can create a local shell, with its own private variables by enclosing commands in parentheses.
#!/bin/csh cd /etc ( cd /usr/bin; ls * ) > myfile pwd
This program changes the working directory to /etc and then executes a subshell which inside the brackets changes directory to /usr/bin and lists the files there. The output of this private shell are sent to a file `myfile'. At the end we print out the current working directory just to show that the `cd' command in brackets had no effect on the main program.
Normally both parentheses must be on the same line. If a subshell command line gets too long, so that the brackets are not on the same line, you have to use backslash characters to continue the lines,
( command \ command \ command \ )
No programming language would be complete without tests and loops. C-shell has two kinds of decision structure: the `if..then..else' and the `switch' structure. These are closely related to their C counterparts. The syntax of these is
if (condition) command if (condition) then command command.. else command command.. endif switch (string) case one: commands breaksw case two: commands breaksw ... endsw
In the latter case, no commands should appear on the same line as a `case' statement, or they will be ignored. Also, if the `breaksw' commands are omitted, then control flows through all the commands for case 2, case 3 etc, exactly as it does in the C programming language.
We shall consider some examples of these statements in a moment, but first it is worth listing some important tests which can be used in `if' questions to find out information about files.
We shall also have need of the following comparison operators.
The simplest way to learn about these statements is to use them, so we shall now look at some examples.
#!/bin/csh -f # # Safe copy from <arg[1]> to <arg[2]> # # if ($#argv != 2) then echo "Syntax: copy <from-file> <to-file>" exit 0 endif if ( -f $argv[2] ) then echo "File exists. Copy anyway?" switch ( $< ) # Get a line from user case y: breaksw default: echo "Doing nothing!" exit 0 endsw endif echo -n "Copying $argv[1] to $argv[2]..." cp $argv[1] $argv[2] echo done endif
This script tries to copy a file from one location to another. If the user does not type exactly two arguments, the script quits with a message about the correct syntax. Otherwise it tests to see whether a plain file has the same name as the file the user wanted to copy to. If such a file exists, it asks the user if he/she wants to continue before proceeding to copy.
Here is another example which compiles a software package. This is a problem we shall return to later See section Make. The problem this script tries to address is the following. There are many different versions of UNIX and they are not exactly compatible with one another. The program this file compiles has to work on any kind of UNIX, so it tries first to determine what kind of UNIX system the script is being run on by calling `uname'. Then it defines a variable `MAKE' which contains the path to the `make' program which will build software. The make program reads a file called `Makefile' which contains instructions for compiling the program, but this file needs to know the type of UNIX, so the script first copies a file `Makefile.src' using `sed' replace a dummy string with the real name of the UNIX. Then it calls make and sets the correct permission on the file using `chmod'.
#!/bin/csh -f ################################################# # # # CONFIGURE Makefile AND BUILD software # # ################################################# set NAME = ( `uname -r -s` ) switch ($NAME[1]) case SunOS*: switch ($NAME[2]) case 4*: setenv TYPE SUN4 setenv MAKE /bin/make breaksw case 5*: setenv TYPE SOLARIS setenv MAKE /usr/ccs/bin/make breaksw endsw breaksw case ULTRIX*: setenv TYPE ULTRIX setenv MAKE /bin/make breaksw case HP-UX*: setenv TYPE HPUX setenv MAKE /bin/make breaksw case AIX*: setenv TYPE AIX setenv MAKE /bin/make breaksw case OSF*: setenv TYPE OSF setenv MAKE /bin/make breaksw case IRIX*: setenv TYPE IRIX setenv MAKE /bin/make breaksw default: echo Unknown architecture $NAME[1] endsw # Generate Makefile from source file sed s/HOSTTYPE/$TYPE/ Makefile.src > Makefile echo "Making software. Type CTRL-C to abort and edit Makefile" $MAKE software # call make to build program chmod 755 software # set correct protection
The C-shell has three loop structures: `repeat', `while' and `foreach'. We have already seen some examples of the `foreach' loop.
The structure of these loops is as follows
repeat number-of-times command while ( test expression ) commands end foreach control-variable ( list-or-array ) commands end
The commands `break' and `continue' can be used to break out of the loops at any time. Here are some examples.
repeat 2 echo "Yo!" | write mark
This sends the message "Yo!" to mark's terminal twice.
repeat 5 echo `echo "Shutdown time! Log out now" | wall ; sleep 30` ; halt
This example repeats the command `echo Shutdown time...' five times at 30 second intervals, before shutting down the system. Only the superuser can run this command! Note the strange construction with `echo echo'. This is to force the repeat command to take two shell commands as an argument. (Try to explain why this works for yourself.)
# Test a user response echo "Answer y/n (yes or no)" set valid = false while ( $valid == false ) switch ( $< ) case y: echo "You answered yes" set valid = true breaksw case n: echo "You answered no" set valid = true breaksw default: echo "Invalid response, try again" breaksw endsw end
Notice that it would have been simpler to replace the two lines
set valid = true breaksw
by a single line `break'. `breaksw' jumps out of the switch construction, after which the `while' test fails. `break' jumps out of the entire while loop.
A path name consists of a number of different parts:
By using one of the following modifiers, we can extract these different elements.
Here are some examples and the results:
set f = ~/progs/c++/test.C echo $f:h /home/mark/progs/c++ echo $f:t test.C echo $f:e C echo $f:r /home/mark/progs/c++/test
Before using these features in a real script, we need one more possibility: numerical addition, subtraction and multiplication etc.
To tell the C-shell that you want to perform an operation on numbers rather than strings, you use the `@' symbol followed by a space. Then the following operations are possible.
@ var = 45 # Assign a numerical value to var echo $var # Print the value @ var = $var + 34 # Add 34 to var @ var += 34 # Add 34 to var @ var -= 1 # subtract 1 from var @ var *= 5 # Multiply var by 5 @ var /= 3 # Divide var by 3 (integer division) @ var %= 3 # Remainder after dividing var by 3 @ var++ # Increment var by 1 @ var-- # Decrement var by 1 @ array[1] = 5 # Numerical array @ logic = ( $x > 6 && $x < 10) # AND @ logic = ( $x > 6 || $x < 10) # OR @ false = ! $var # Logical NOT @ bits = ( $x | $y ) # Bitwise OR @ bits = ( $x ^ $y ) # Bitwise XOR @ bits = ( $x & $y ) # Bitwise AND @ shifted = ( $var >> 2 ) # Bitwise shift right @ back = ( $var << 2 ) # Bitwise shift left
These operators are precisely those found in the C programming language.
The following script uses the operators in the last two sections to take a list of files with a given file extension (say `.doc') and change it for another (say `.tex'). This is a partial solution to the limitation of not being able to do multiple renames in shell.
#!/bin/csh -f ############################################################# # # Change file extension for multiple files # ############################################################# if ($#argv < 2) then echo Syntax: chext oldpattern newextension echo "e.g: chext *.doc tex " exit 0 endif mkdir /tmp/chext.$user # Make a scratch area set newext="$argv[$#argv]" # Last arg is new ext set oldext="$argv[1]:e" echo "Old extension was ($oldext)"" echo "New extension ($newext) -- okay? (y/n)" switch( $< ) case y: breaksw default: echo "Nothing done." exit 0 endsw ############################################################## # Remove the last file extension from files ############################################################## i = 0 foreach file ($argv) i++ if ( $i == $#argv ) break cp $file /tmp/chext.$user/$file:r # temporary store end ############################################################### # Add .newext file extension to files ############################################################### set array = (`ls /tmp/chext.$user`) foreach file ($array) if ( -f $file.$newext ) then echo destination file $file.$newext exists. No action taken. continue endif cp /tmp/chext.$user/$file $file.$newext rm $file.$oldext end rm -r /tmp/chext.$user
Here is another example to try to decipher. Use the manual pages to find out about `awk'. This script can be written much more easily in Perl or C, as we shall see in the next chapters. It is also trivially implemented as a script in the system administration language cfengine.
#!/bin/csh -f ########################################################### # # KILL all processes owned by $argv[1] with PID > $argv[2] # ########################################################### if ("`whoami`" != "root") then echo Permission denied exit 0 endif if ( $#argv < 1 || $#argv > 2 ) then echo Usage: KILL username lowest-pid exit 0 endif if ( $argv[1] == "root") then echo No! Too dangerous -- system will crash exit 0 endif ############################################################ # Kill everything ############################################################ if ( $#argv == 1 ) then set killarray = ( `ps aux | awk '{ if ($1 == user) \ {printf "%s ",$2}}' user=$argv[1]` ) foreach process ($killarray) kill -1 $process kill -15 $process > /dev/null kill -9 $process > /dev/null if ("`kill -9 $process | egrep -e 'No such process'`" == "") then echo "Warning - $process would not die - try again" endif end ############################################################# # Start from a certain PID ############################################################# else if ( $#argv == 2 ) then set killarray = ( `ps aux | awk '{ if ($1 == user && $2 > uid) \ {printf "%s ",$2}}' user=$argv[1] uid=$argv[2]` ) foreach process ($killarray) kill -1 $process > /dev/null kill -15 $process sleep 2 kill -9 $process > /dev/null if ("`kill -9 $process | egrep -e 'No such process'`" == "") then echo "Warning - $process would not die - try again" endif end endif
This program would be better written in C or Perl.
To summarize the last two long and oppressive chapters we shall take a step back from the details and look at what we have achieved.
The idea behind the shell is to provide a user interface, with access to the system's facilities at a simple level. In the 70's user interfaces were not designed to be user-friendly. The UNIX shell is not particularly use friendly, but it is very powerful. Perhaps it would have been enough to provide only commands to allow users to write C programs. Since all of the system functions are available from C, that would certainly allow everyone to do what anything that UNIX can do. But shell programming is much more immediate than C. It is an environment of frequently used tools. Also for quick programming solutions: C is a compiled language, whereas the shell is an interpreter. A quick shell program can solve many problems in no time at all, without having to compile anything.
Shell programming is only useful for `quick and easy' programs. To use it for anything serious is an abuse. Programming difficult things in shell is clumsy, and it is difficult to get returned-information (like error messages) back in a useful form. Besides, shell scripts are slow compared to real programs since they involve starting a new program for each new command.
These difficulties are solved partly by Perl, which we shall consider next -- but in the final analysis, real programs of substance need to be written in C. Contrary to popular belief, this is not more difficult than programming in the shell -- in fact, many things are much simpler, because all of the shell commands originated as C functions. The shell is an extra layer of the UNIX onion which we have to battle our way through to get where we're going.
Sometimes it is helpful to be shielded from low level details -- sometimes it is a hindrance. In the remaining chapters we shall consider more involved programming needs.
So far, we have been looking at shell programming for performing fairly simple tasks. Now let's extend the idea of shell programming to cover more complex tasks like systems programming and network communications. Perl is a language which was designed to retain the immediateness of shell languages, but at the same time capture some of the flexibility of C. Perl is an acronym for Practical extraction and report language. In this chapter, we shall not aim to teach Perl from scratch -- the best way to learn it is to use it! Rather we shall concentrate on demonstrating some principles.
One of the reasons for using Perl is that it is extremely good at textfile handling--one of the most important things for UNIX users, and particularly useful in connection with CGI script processing on the World Wide Web. It has simple built-in constructs for searching and replacing text, storing information in arrays and retrieving them in sorted form. All of the these things have previously been possible using the UNIX shell commands
sed awk cut paste
but these commands were designed to work primarily in the Bourne shell and are a bit `awk'ward to use for all but the simplest applications.
stdin
and produces output on stdout according
to those instructions. `sed' works line by line from the start of a
textfile.
Perl unifies all of these operations and more. It also makes them much simpler.
To summarize Perl, we need to know about the structure of a Perl program, the conditional constructs it has, its loops and its variables. In the latest versions of Perl (Perl 5), you can write object oriented programs of great complexity. We shall not go into this depth, for the simple reason that Perl's strength is not as a general programming language but as a specialized language for textfile handling. The syntax of Perl is in many ways like the C programming language, but there are important differences.
`command`
can be used to execute UNIX
programs and get the result into a Perl variable.
Here is a simple `structured hello world' program in Perl. Notice that subroutines are called using the `&' symbol. There is no special way of marking the main program -- it is simply that part of the program which starts at line 1.
#!/local/bin/perl # # Comments # &Hello(); &World; # end of main sub Hello { print "Hello"; } sub World { print "World\n"; }
The parentheses on subroutines are optional, if there are no parameters passed. Notice that each line must end in a semi-colon.
In Perl, variables do not have to be declared before they are used. Whenever you use a new symbol, Perl automatically adds the symbol to its symbol table and initializes the variable to the empty string.
It is important to understand that there is no practical difference between zero and the empty string in perl -- except in the way that you, the user, choose to use it. Perl makes no distinction between strings and integers or any other types of data -- except when it wants to interpret them. For instance, to compare two variables as strings is not the same as comparing them as integers, even if the string contains a textual representation of an integer. Take a look at the following program.
#!/local/bin/perl # # Nothing! # print "Nothing == $nothing\n"; print "Nothing is zero!\n" if ($nothing == 0); if ($nothing eq "") { print STDERR "Nothing is really nothing!\n"; } $nothing = 0; print "Nothing is now $nothing\n";
The output from this program is
Nothing == Nothing is zero! Nothing is really nothing! Nothing is now 0
There are several important things to note here. First of all, we never declare the variable `nothing'. When we try to write its value, perl creates the name and associates a NULL value to it i.e. the empty string. There is no error. Perl knows it is a variable because of the `$' symbol in front of it. All scalar variables are identified by using the dollar symbol.
Next, we compare the value of `$nothing' to the integer `0' using the integer comparison symbol `==', and then we compare it to the empty string using the string comparison symbol `eq'. Both tests are true! That means that the empty string is interpreted as having a numerical value of zero. In fact any string which does not form a valid integer number has a numerical value of zero.
Finally we can set `$nothing' explicitly to a valid integer string zero, which would now pass the first test, but fail the second.
As extra spice, this program also demonstrates two different ways of writing the `if' command in perl.
The special variable `$_' is used for many purposes in Perl. It is used as a buffer to contain the result of the last operation, the last line read in from a file etc. It is so general that many functions which act on scalar variables work by default on `$_' if no other argument is specified. For example,
print;
is the same as
print $_;
The complement of scalar variables is arrays. An array, in Perl is identified by the `@' symbol and, like scalar variables, is allocated and initialized dynamically.
@array[0] = "This little piggy went to market"; @array[2] = "This little piggy stayed at home"; print "@array[0] @array[1] @array[2]";
The index of an array is always understood to be a number, not a string, so if you use a non-numerical string to refer to an array element, you will always get the zeroth element, since a non-numerical string has an integer value of zero.
An important array which every program defines is
@ARGV
This is the argument vector array, and contains the commands line arguments by analogy with the C-shell variable `$argv[]'.
Given an array, we can find the last element by using the `$#' operator. For example,
$last_element = $ARGV[$#ARGV];
Notice that each element in an array is a scalar variable. The `$#' cannot be interpreted directly as the number of elements in the array, as it can in the C-shell. You should experiment with the value of this quantity -- it often necessary to add 1 or 2 to its value in order to get the behaviour one is used to in the C-shell.
Perl does not support multiple-dimension arrays directly, but it is possible to simulate them yourself. (See the Perl book.)
The `shift' command acts on arrays and returns and removes the first element of the array. Afterwards, all of the elements are shifted down one place. So one way to read the elements of an array in order is to repeatedly call `shift'.
$next_element=shift(@myarray);
Note that, if the array argument is omitted, then `shift' works on `@ARGV' by default.
Another useful function is `split', which takes a string and turns it into an array of strings. `split' works by choosing a character (usually a space) to delimit the array elements, so a string containing a sentence separated by spaces would be turned into an array of words. The syntax is
@array = split; # works with spaces on $_ @array = split(pattern,string); # Breaks on pattern ($v1,$v2...) = split(pattern,string); # Name array elements with scalars
In the first of these cases, it is assumed that the variable `$_' is to be split on whitespace characters. In the second case, we decide on what character the split is to take place and on what string the function is to act. For instance
@new_array = split(":","name:passwd:uid:gid:gcos:home:shell");
The result is a seven element array called `@new_array', where `$new_array[0]' is `name' etc.
In the final example, the left hand side shows that we wish to capture elements of the array in a named set of scalar variables. If the number of variables on the lefthand side is fewer than the number of strings which are generated on the right hand side, they are discarded. If the number on the left hand side is greater, then the remainder variables are empty.
One of the very nice features of Perl is the ability to use one string as an index to another string in an array. For example, we can make a short encyclopedia of zoo animals by constructing an associative array in which the keys (or indices) of the array are the names of animals, and the contents of the array are the information about them.
$animals{"Penguin"} = "A suspicious animal, good with cheese crackers..."; $animals{"dog"} = "Plays stupid, but could be a cover..."; if ($index eq "fish") { $animals{$index} = "Often comes in square boxes. Very cold."; }
An entire associated array is written `%array', while the elements are `$array{$key}'.
Perl provides a special associative array for every program called `%ENV'. This contains the environment variables defined in the parent shell which is running the Perl program. For example
print "Username = $ENV{"USER"}\n"; $ld = "LD_LIBRARY_PATH"; print "The link editor path is $ENV{$ld}\n";
To get the current path into an ordinary array, one could write,
@path_array= split(":",$ENV{"PATH"});
Here is an example which prints out a list of files in a specified directory, in order of their UNIX protection bits. The least protected file files come first.
#!/local/bin/perl # # Demonstration of arrays and associated arrays. # Print out a list of files, sorted by protection, # so that the least secure files come first. # # e.g. arrays <list of words> # arrays *.C # ############################################################ print "You typed in ",$#ARGV+1," arguments to command\n"; if ($#ARGV < 1) { print "That's not enough to do anything with!\n"; } while ($next_arg = shift(@ARGV)) { if ( ! ( -f $next_arg || -d $next_arg)) { print "No such file: $next_arg\n"; next; } ($dev,$ino,$mode,$nlink,$uid,$gid,$rdev,$size) = stat($next_arg); $octalmode = sprintf("%o",$mode & 0777); $assoc_array{$octalmode} .= $next_arg. " : size (".$size."), mode (".$octalmode.")\n"; } print "In order: LEAST secure first!\n\n"; foreach $i (reverse sort keys(%assoc_array)) { print $assoc_array{$i}; }
Here are some of the most commonly used decision-making constructions and loops in Perl. The following is not a comprehensive list -- for that, you will have to look in the Perl bible: Programming Perl, by Larry Wall and Randal Schwartz. The basic pattern follows the C programming language quite closely. In the case of the `for' loop, Perl has both the C-like version, called `for' and a `foreach' command which is like the C-shell implementation.
if (expression) { block; } else { block; } command if (expression); unless (expression) { block; } else { block; } while (expression) { block; } do { block; } while (expression); for (initializer; expression; statement) { block; } foreach variable(array) { block; }
In all cases, the `else' clauses may be omitted.
Strangely, perl does not have a `switch' statement, but the Perl book describes how to make one using the features provided.
The for loop is exactly like that in C or C++ and is used to iterate over a numerical index, like this:
for ($i = 0; $i < 10; $i++) { print $i, "\n"; }
The foreach loop is like its counterpart in the C shell. It is used for reading elements one by one from a regular array. For example,
foreach $i ( @array ) { print $i, "\n"; }
One of the main uses for `for' type loops is to iterate
over successive values in an array. This can be done in
two ways which show the essential difference between
for
and foreach
.
If we want to fetch each value in an array in turn, without caring about
numerical indices, the it is simplest to use the foreach
loop.
@array = split(" ","a b c d e f g"); foreach $var ( @array ) { print $var, "\n"; }
This example prints each letter on a separate line.
If, on the other hand, we are interested in the index,
for the purposes of some calculation, then the for
loop is preferable.
@array = split(" ","a b c d e f g"); for ($i = 0; $i <= $#array; $i++) { print $array[$i], "\n"; }
Notice that, unlike the for-loop idiom in C/C++, the limit is `$i <= $#array', i.e. `less than or equal to' rather than `less than'. This is because the `$#' operator does not return the number of elements in the array but rather the last element.
Associated arrays are slightly different, since they do not
use numerical keys. Instead they use a set of strings,
like in a database, so that you can use one string to look
up another. In order to iterate over the values in the array
we need to get a list of these strings. The keys
command is used for this.
$assoc{"mark"} = "cool"; $assoc{"GNU"} = "brave"; $assoc{"zebra"} = "stripy"; foreach $var ( keys %assoc ) { print "$var , $assoc{$var} \n"; }
The order of the keys is not defined in the above example, but you can choose to sort them alphabetically by writing
foreach $var ( sort keys %assoc )
instead.
Since Perl is about file handling we are very interested in reading files. Unlike C and C++, perl likes to read files line by line. The angle brackets are used for this, See section Files in perl. Assuming that we have some file handle `<file>', for instance `<STDIN>', we can always read the file line by line with a while-loop like this.
while ($line = <file>) { print $line; }
Note that $line
includes the end of line character on the end
of each line. If you want to remove it, you should add a `chop'
command:
while ($line = <file>) { chop $line; print "line = ($line)\n"; }
Opening files is straightforward in Perl. Files must be opened and closed using -- wait for it -- the commands `open' and `close'. You should be careful to close files after you have finished with them -- especially if you are writing to a file. Files are buffered and often large parts of a file are not actually written until the `close' command is received.
Three files are, of course, always open for every program, namely `STDIN', `STDOUT'and `STDERR'.
Formally, to open a file, we must obtain a file descriptor or file handle. This is done using `open';
open (file_descrip,"Filename");
The angular brackets `<..>' are used to read from the file. For example,
$line = <file_descrip>;
reads one line from the file associated with `file_descrip'.
Let's look at some examples of filing opening. Here is how we can implement UNIX's `cut' and `paste' commands in perl:
#!/local/bin/perl # # Cut in perl # # Cut second column while (<>) { @cut_array = split; print "@cut_array[1]\n"; }
This is the simplest way to open a file. The empty file descriptor `<>' tells perl to take the argument of the command as a filename and open that file for reading. This is really short for `while($_=<STDIN>)' with the standard input redirected to the named file.
The `paste'program can be written as follows:
#!/local/bin/perl # # Paste in perl # # Two files only, syntax : paste file 1file2 # open (file1,"@ARGV[0]") || die "Can't open @ARGV[0]\n"; open (file2,"@ARGV[1]") || die "Can't open @ARGV[1]\n"; while (($line1 = <file1>) || ($line2 = <file2>)) { chop $line1; chop $line2; print "$line1 $line2\n"; # tab character between }
Here we see more formally how to read from two separate files at the same time. Notice that, by putting the read commands into the test-expression for the `while' loop, we are using the fact that `<..>' returns a non-zero (true) value unless we have reached the end of the file.
To write and append to files, we use the shell redirection symbols inside the `open' command.
open(fd,"> filename"); # open file for writing open(fd,">> filename"); # open file for appending
We can also open a pipe from an arbitrary UNIX command and receive the output of that command as our input:
open (fd,"/bin/ps aux | ");
Let us now write the simplest perl program which illustrates the way in which perl can save time. We shall write it in three different ways to show what the short cuts mean. Let us implement the `cat' command, which copies files to the standard output. The simplest way to write this is perl is the following:
#!/local/bin/perl while (<>) { print; }
Here we have made heavy use of the many default assumptions which perl
makes. The program is simple, but difficult to understand for
novices. First of all we use the default file handle <>
which
means, take one line of input from a default file. This object returns
true as long as it has not reached the end of the file, so this loop
continues to read lines until it reaches the end of file. The default
file is standard input, unless this script is invoked with a command
line argument, in which case the argument is treated as a filename and
perl attempts to open the argument-filename for reading. The
print
statement has no argument telling it what to print, but
perl takes this to mean: print the default variable `$_'.
We can therefore write this more explicitly as follows:
#!/local/bin/perl open (HANDLE,"$ARGV[1]"); while (<HANDLE>) { print $_; }
Here we have simply filled in the assumptions explicitly. The command `<HANDLE>' now reads a single line from the named file-handle into the default variable `$_'. To make this program more general, we can eliminate the defaults entirely.
#!/local/bin/perl open (HANDLE,"$ARGV[1]"); while ($line=<HANDLE>) { print $line; }
Be careful to distinguish between the comparison operator for integers `==' and the corresponding operator for strings `eq'. These do not work in each other's places so if you get the wrong comparison operator your program might not work and it is quite difficult to find the error.
The command `chop' cuts off the last character of a string. This is useful for removing newline characters when reading files etc. The syntax is
chop; # chop $_; chop $scalar; # remove last character in $scalar
Subroutines are indicated, as in the example above, by the ampersand `&' symbol. When parameters are passed to a Perl subroutine, they are handed over as an array called `@_'. Which is analogous to the `$_' variable. Here is a simple example:
#!/local/bin/perl $a="silver"; $b="gold"; &PrintArgs($a,$b); # end of main sub PrintArgs { ($local_a,$local_b) = @_; print "$local_a, $local_b\n"; }
When a program has to quit and give a message, the `die' command is normally used. If called without an argument, Perl generates its own message including a line number at which the error occurred. To include your own message, you write
die "My message....";
If the string is terminated with a `\n' newline character, the line number of the error is not printed, otherwise Perl appends the line number to your string.
When opening files, it is common to see the syntax:
open (filehandle,"Filename") || die "Can't open...";
The logical `OR' symbol is used, because `open' returns true if all goes well, in which case the right hand side is never evaluated. If `open' is false, then die is executed. You can decide for yourself whether or not you think this is good programming style -- we mention it here because it is common practice.
stat()
idiom
The UNIX library function stat()
is used to find out information
about a given file. This function is available both in C and in Perl.
In perl, it returns an array of values. Usually we are interested in
knowing the access permissions of a file. stat()
is called
using the syntax
@array = stat ("filename");
or alternatively, using a named array
($device,$inode,$mode) = stat("filename");
The value returned in the mode variable is a bit-pattern, See section Protection bits. The most useful way of treating these bit patterns is to use octal numbers to interpret their meaning.
To find out whether a file is readable or writable to a group of users, we use a programming idiom which is very common for dealing with bit patterns: first we define a mask which zeroes out all of the bits in the mode string except those which we are specifically interested in. This is done by defining a mask value in which the bits we want are set to 1 and all others are set to zero. Then we AND the mask with the mode string. If the result is different from zero then we know that all of the bits were also set in the mode string. As in C, the bitwise AND operator in perl is called `&'.
For example, to test whether a file is writable to other users in the same group as the file, we would write the following.
$mask = 020; # Leading 0 means octal number ($device,$inode,$mode) = stat("file"); if ($mode & $mask) { print "File is writable by the group\n"; }
Here the 2 in the second octal number means "write", the fact that it is the second octal number from the right means that it refers to "group". Thus the result of the if-test is only true if that particular bit is true. We shall see this idiom in action below.
Here is a simple implementation of the UNIX `passwd' program in Perl.
#!/local/bin/perl # # A perl version of the passwd program. # # Note - the real passwd program needs to be much more # secure than this one. This is just to demonstrate the # use of the crypt() function. # ############################################################# print "Changing passwd for $ENV{'USER'} on $ENV{'HOST'}\n"; system 'stty','-echo'; print "Old passwd: "; $oldpwd = <STDIN>; chop $oldpwd; ($name,$coded_pwd,$uid,$gid,$x,$y,$z,$gcos,$home,$shell) = getpwnam($ENV{"USER"}); if (crypt($oldpwd,$coded_pwd) ne $coded_pwd) { print "\nPasswd incorrect\n"; exit (1); } $oldpwd = ""; # Destroy the evidence! print "\nNew passwd: "; $newpwd = <STDIN>; print "\nRepeat new passwd: "; $rnewpwd = <STDIN>; chop $newpwd; chop $rnewpwd; if ($newpwd ne $rnewpwd) { print "\n Incorrectly typed. Password unchanged.\n"; exit (1); } $salt = rand(); $new_coded_pwd = crypt($newpwd,$salt); print "\n\n$name:$new_coded_pwd:$uid:$gid:$gcos:$home:$shell\n";
The following example uses the `fork' function to start a daemon which goes into the background and watches the system to which process is using the greatest amount of CPU time each minute. A pipe is opened from the BSD `ps' command.
#!/local/bin/perl # # A fork() demo. This program will sit in the background and # make a list of the process which uses the maximum CPU average # at 1 minute intervals. On a quiet BSD like system this will # normally be the swapper (long term scheduler). # $true = 1; $logfile="perl.cpu.logfile"; print "Max CPU logfile, forking daemon...\n"; if (fork()) { exit(0); } while ($true) { open (logfile,">> $logfile") || die "Can't open $logfile\n"; open (ps,"/bin/ps aux |") || die "Couldn't open a pipe from ps !!\n"; $skip_first_line = <ps>; $max_process = <ps>; close(ps); print logfile $max_process; close(logfile); sleep 60; ($a,$b,$c,$d,$e,$f,$g,$size) = stat($logfile); if ($size > 500) { print STDERR "Log file getting big, better quit!\n"; exit(0); } }
Here is an example program with several of the above features demonstrated simultaneously. This following program lists all users who have home directories on the current host. If the home area has sub-directories, corresponding to groups, then this is specified on the command line. The word `home' causes the program to print out the home directories of the users.
#!/local/bin/perl ################################################################## # # allusers - list all users on named host, i.e. all # users who can log into this machine. # # Syntax: allusers group # allusers mygroup home # allusers myhost group home # # NOTE : This command returns only users who are registered on # the current host. It will not find users which cannot # be validated in the passwd file, or in the named groups # in NIS. It assumes that the users belonging to # different groups are saved in subdirectories of # /home/hostname. # ################################################################## &arguments(); die "\n" if ( ! -d "/home/$server" ); $disks = `/bin/ls -d /home/$server/$group`; foreach $home (split(/\s/,$disks)) { open (LS,"cd $home; /bin/ls $home |") || die "allusers: Pipe didn't open"; while (<LS>) { $exists = ""; ($user) = split; ($exists,$pw,$uid,$gid,$qu,$cm,$gcos,$dir)=getpwnam($user); if ($exists) { if ($printhomes) { print "$dir\n"; } else { print "$user\n"; } } } close(LS); } ######################################################## sub arguments { $printhomes = 0; $group = "*"; $server = `/bin/hostname`; chop $server; foreach $arg (@ARGV) { if (substr($arg,0,1) eq "u") { $group = $arg; next; } if ($arg eq "home") { $printhomes = 1; next; } $server= $arg; #default is to interpret as a server. } }
Perl has regular expression operators for identifying patterns. The operator
/regular expression/
returns true of false depending on whether the regular expression matches the
contents of $_
. For example
if (/perl/) { print "String contains perl as a substring"; } if (/(Sat|Sun)day/) { print "Weekend day...."; }
The effect is rather like the grep
command.
To use this operator on other variables you would write:
$variable =~ /regexp/
Regular expression can contain parenthetic sub-expressions, e.g.
if (/(Sat|Sun)day (..)th (.*)/) { $first = $1; $second = $2; $third = $3; }
in which case perl places the objects matched by such sub-expressions
in the variables $1
, $2
etc.
The `sed'-like function for replacing all occurances of a string is easily implemented in Perl using
while (<input>) { s/$search/$replace/g; print output; }
This example replaces the string inside the default variable. To replace in a general variable we use the operator `=~', with syntax:
$variable =~ s/search/replace/
Here is an example of some of this operator in use. The following is a program which searches and replaces a string in several files. This is useful program indeed for making a change globally in a group of files! The program is called `file-replace'.
#!/local/bin/perl ############################################################## # # Look through files for findstring and change to newstring # in all files. # ############################################################## # # Define a temporary file and check it doesn't exist # $outputfile = "tmpmarkfind"; unlink $outputfile; # # Check command line for list of files # if ($#ARGV < 0) { die "Syntax: file-replace [file list]\n"; } print "Enter the string you want to find (Don't use quotes):\n\n:"; $findstring=<STDIN>; chop $findstring; print "Enter the string you want to replace with (Don't use quotes):\n\n:"; $replacestring=<STDIN>; chop $replacestring; # print "\nFind: $findstring\n"; print "Replace: $replacestring\n"; print "\nConfirm (y/n) "; $y = <STDIN>; chop $y; if ( $y ne "y") { die "Aborted -- nothing done.\n"; } else { print "Use CTRL-C to interrupt...\n"; } # # Now shift default array @ARGV to get arguments 1 by 1 # while ($file = shift) { if ($file eq "file-replace") { print "Findmark will not operate on itself!"; next; } # # Save existing mode of file for later # ($dev,$ino,$mode)=stat($file); open (INPUT,$file) || warn "Couldn't open $file\n"; open (OUTPUT,"> $outputfile") || warn "Can't open tmp"; $notify = 1; while (<INPUT>) { if (/$findstring/ && $notify) { print "Fixing $file...\n"; $notify = 0; } s/$findstring/$replacestring/g; print OUTPUT; } close (OUTPUT); # # If nothing went wrong (if outfile not empty) # move temp file to original and reset the # file mode saved above # if (! -z $outputfile) { rename ($outputfile,$file); chmod ($mode,$file); } else { print "Warning: file empty!\n."; } }
Similarly we can search for lines containing a string. Here is the grep program written in perl
#!/local/bin/perl # # grep as a perl program # # Check arguments etc while (<>) { print if (/$ARGV[1]/); }
The operator `/search-string/' returns true if the search
string is a substring of the default variable $_
. To search
an arbitrary string, we write
.... if (teststring =~ /search-string/);
Here teststring is searched for occurrances of search-string and the result is true if one is found.
In perl you can use regular expressions to search for text patterns. Note however that, like all regular expression dialects, perl has its own conventions. For example the dollar sign does not mean "match the end of line" in perl, instead one uses the `\n' symbol. Here is an example program which illustrates the use of regular expressions in perl:
#!/local/bin/perl # # Test regular expressions in perl # # NB - careful with \ $ * symbols etc. Use " quotes since # the shell interprets these! # open (FILE,"regex_test"); $regex = $ARGV[$#ARGV]; print "Looking for $ARGV[$#ARGV] in file...\n"; while (<FILE>) { if (/$regex/) { print; } } # # Test like this: # # regex '.*' - prints every line (matches everything) # regex '.' - all lines except those containing only blanks # (. doesn't match ws/white-space) # regex '[a-z]' - matches any line containing lowercase # regex '[^a-z]' - matches any line containg something which is # not lowercase a-z # regex '[A-Za-z]' - matches any line containing letters of any kind # regex '[0-9]' - match any line containing numbers # regex '#.*' - line containing a hash symbol followed by anything # regex '^#.*' - line starting with hash symbol (first char) # regex ';\n' - match line ending in a semi-colon #
Try running this program with the test data on the following file which is called `regex_test' in the example program.
# A line beginning with a hash symbol JUST UPPERCASE LETTERS just lowercase letters Letters and numbers 123456 123456 A line ending with a semi-colon; Line with a comment # COMMENT...
Here is an example program which you could use to automatically turn a mail message of the form
From: Newswire To: Mail2html Subject: Nothing happened On the 13th February at kl. 09:30 nothing happened. No footprints were found leading to the scene of a terrible murder, no evidence of a struggle .... etc etc
into an html-file for the world wide web. The program works by extracting the message body and subject from the mail and writing html-commands around these to make a web page. The subject field of the mail becomes the title. The other headers get skipped, since the script searches for lines containing the sequence "colon-space" or `: '. A regular expression is used for this.
#!/local/bin/perl # # Make HTML from mail # &BeginWebPage(); &ReadNewMail(); &EndWebPage(); ########################################################## sub BeginWebPage { print "<HTML>\n"; print "<BODY>\n"; } ########################################################## sub EndWebPage { print "</BODY>\n"; print "</HTML>\n"; } ########################################################## sub ReadNewMail { while (<>) { if (/Subject:/) # Search for subject line { # Extract subject text... chop; ($left,$right) = split(":",$_); print "<H1> $right </H1>\n"; next; } elsif (/.*: .*/) # Search for - anything: anything { next; # skip other headers } print; } }
The following program scans through the password database and
build a standardized html-page for each user it finds there. It
fills in the name of the user in each case. Note the use of
the `<<' operator for extended input, already used in the
context of the shell, See section Pipes and redirection in csh. This allows
us to format a whole passage of text, inserting variables at
strategic places, and avoid having to the print
over
many lines.
#!/local/bin/perl # # Build a default home page for each user in /etc/passwd # # #################################################################### # Level 0 (main) #################################################################### $true = 1; $false = 0; # First build an associated array of users and full names setpwent(); while ($true) { ($name,$passwd,$uid,$gid,$quota,$comment,$fullname) = getpwent; $FullName{$name} = $fullname; print "$name - $FullName{$name}\n"; last if ($name eq ""); } print "\n"; # Now make a unique filename for each page and open a file foreach $user (sort keys(%FullName)) { next if ($user eq ""); print "Making page for $user\n"; $outputfile = "$user.html"; open (OUT,"> $outputfile") || die "Can't open $outputfile\n"; &MakePage; close (OUT); } #################################################################### # Level 1 #################################################################### sub MakePage { print OUT <<ENDMARKER; <HTML> <BODY> <HEAD><TITLE>$FullName{$user}'s Home Page</TITLE></HEAD> <H1>$FullName{$user}'s Home Page</H1> Hi welcome to my home page. In case you hadn't got it yet my name is: $FullName{$user}... I study at <a href=http://www.iu.hio.no>Høgskolen i Oslo</a>. </BODY> </HTML> ENDMARKER }
Perl has very many functions which come directly from the C library. To give a taster, a few are listed here. The Perl book contains a comprehensive description of these.
Here are some of the most frequently used functions
chmod
chmod 755,filename
chdir
chdir /etc
stat
open
close
system
system "ls";
split
@array=split(":",$string)
.
rename
rename old name new-name
mkdir
mkdir newdir
shift
$first=shift(@array);)
.
chop
oct
$decimal = oct(755);
kill
kill -9, pid1,pid2...
You should explore Perl's possibilities yourself. Perl is a good alternative to the shell which has much of the power of C and is therefore ideal for simple and more complex system programming tasks. If you intend to be a system administrator for UNIX systems, you could do much worse than to read the Perl book and learn Perl inside out.
The Practical Extraction and Report Language is a powerful tool which goes beyond shell programming, but which retains much of the immediateness of shell programming in a more formal programming environment.
The success of Perl has led many programmers to use it exclusively. In the next section, I would like to argue that programming directly in C is not much harder. In fact it has advantages in the long run. The power of Perl is that it is as immediate as shell programming. If you are inexperienced, Perl is a little easier than C because many features are ready programmed into the language, but with time one also builds up a repertoire of C functions which can do the same tricks.
Write a program which checks the `sanity' of your UNIX system.
CGI stands for the Common Gateway Interface. It is the name given to scripts which can be executed from within pages of the world wide web. Although it is possible to use any language in CGI programs (hence the word `common'), the usual choice is Perl, because of the ease with which Perl can handle text.
The CGI interface is pretty unintelligent, in order to be as general as possible, so we need to do a bit of work in order to make scripts work.
The key thing about the WWW which often causes a lot of confusion is that the W3 service runs with a user ID of `nobody'. The purpose of this is to ensure that nobody has the right to read or write files unless they are opened very explicitly by the user who owns them.
In order for files to be readable on the WWW, they must have file mode `644' and they must lie in a directory which has mode `755'. In order for a CGI program to be executable, it must have permission `755' and in order for such a program to write to a file in a user's directory, it must be possible for the file to be created (if necessary) and everyone must be able to write to it. That means that files which are written to by the WWW must have mode `666' and must either exist already or lie in a directory with permission `777'(6).
CGI script programs communicate with W3 browsers using a very simple protocol. It goes like this:
Content-type: text/htmlThis must be followed by a blank line.
To start a CGI program from a web page we use a form which is a part of the HTML code enclosed with the parentheses
<FORM method="POST" ACTION="/cgi-script-alias/program.pl"> ... </FORM>
The method `post' means that the data which get typed into this form will be piped into the CGI program via its standard input. The `action' specifies which program you want to start. Note that you cannot simply use the absolute path of the file, for security reasons. You must use something called a `script alias' to tell the web browser where to find the program. If you do not have a script alias defined for you personally, then you need to get one from your system administrator. By using a script alias, no one from outside your site can see where your files are located, only that you have a `cgi-bin' area somewhere on your system.
Within these parentheses, you can arrange to collect different kinds of input. The simplest kind of input is just a button which starts the CGI program. This has the form
<INPUT TYPE="submit" VALUE="Start my program">
This code creates a button. When you click on it the program in your `action' string gets started. More generally, you will want to create input boxes where you can type in data. To create a single line input field, you use the following syntax:
<INPUT NAME="variable-name" SIZE=40>
This creates a single line text field of width 40 characters. This is not the limit on the length of the string which can be typed into the field, only a limit on the amount which is visible at any time. It is for visual formatting only. The NAME field is used to identify the data in the CGI script. The string you enter here will be sent to the CGI script in the form `variable-name=value of input...'. Another type of input is a text area. This is a larger box where one can type in text on several lines. The syntax is:
<TEXTAREA NAME="variable-name" ROW=50 COLS=50>
which means: create a text area of fifty rows by fifty columns with a prompt to the left of the box. Again, the size has only to do with the visual formatting, not to do with limits on the amount of text which can be entered.
As an example, let's create a WWW page with a complete form which can be used to make a guest book, or order form.
<HTML> <HEAD> <TITLE>Example form</TITLE> <!-- Comment: Mark Burgess, 27-Jan-1997 --> <LINK REV="made" HREF="mailto:mark@iu.hio.no"> </HEAD> <BODY> <CENTER><H1>Write in my guest book...</H1></CENTER> <HR> <CENTER><H2>Please leave a comment using the form below.</H2><P> <FORM method="POST" ACTION="/cgi-bin-mark/comment.pl"> Your Name/e-mail: <INPUT NAME="variable1" SIZE=40> <BR><BR> <P> <TEXTAREA NAME="variable2" cols=50 rows=8></TEXTAREA> <P> <INPUT TYPE=submit VALUE="Add message to book"> <INPUT TYPE=reset VALUE="Clear message"> </FORM> <P> </BODY> </HTML>
The reset button clears the form. When the submit button is pressed, the CGI program is activated.
To interpret and respond to the data in a form, we must write a program which satisfies the protocol above, See section Protocols. We use perl as a script language. The simplest valid CGI script is the following:
#!/local/bin/perl # # Reply with proper protocol # print "Content-type: text/html\n\n"; # # Get the data from the form ... # $input = <STDIN>; # # ... and echo them back # print $input, "\n Done! \n";
Although rather banal, this script is a useful starting point for CGI programming, because it shows you just how the input arrives at the script from the HTML form. The data arrive all in a single, enormously long line, full of funny characters. The first job of any script is to decode this line.
Before looking at how to decode the data, we should make an important point about the protocol line. If a web browser does not get this `Content-type' line from the CGI script it returns with an error:
500 Server Error The server encountered an internal error or misconfiguration and was unable to complete your request. Please contact the server administrator, and inform them of the time the error occurred, and anything you might have done that may have caused the error. Error: HTTPd: malformed header from script www/cgi-bin/comment.pl
Before finishing your CGI script, you will probably encounter this error several times. A common reason for getting the error is a syntax error in your script. If your program contains an error, the first thing a browser gets in return is not the `Content-type' line, but an error message. The browser does not pass on this error message, it just prints the uninformative message above.
If you can get the above script to work, then you are ready to decode the data which are sent to the script. The first thing is to use perl to split the long line into an array of lines, by splitting on `&'. We can also convert all of the `+' symbols back into spaces. The script now looks like this:
#!/local/bin/perl # # Reply with proper protocol # print "Content-type: text/html\n\n"; # # Get the data from the form ... # $input = <STDIN>; # # ... and echo them back # print "$input\n\n\n"; $input =~ s/\+/ /g; # # Now split the lines and convert # @array = split('&',$input); foreach $var ( @array ) { print "$var\n"; } print "Done! \n";
We now have a series of elements in our array. The output from this script is something like this:
variable1=Mark+Burgess&variable2=%0D%0AI+just+called+to+say+ (wrap) ....%0D%0A...hey+pig%2C+nothing%27s+working+out+the+way+I+planned variable1=Mark Burgess variable2=%0D%0AI just called to say (wrap) ....%0D%0A...hey pig%2Cnothing%27s working out the way I planned Done!
As you can see, all control characters are converted into the form `%XX'. We should now try to do something with these. Since we are usually not interested in keeping new lines, or any other control codes, we can simply null-out these with a line of the form
$input =~ s/%..//g;
The regular expression `%..' matches anything beginning with a percent symbol followed by two characters. The resulting output is then free of these symbols. We can then separate the variable contents from their names by splitting the input. Here is the complete code:
#!/local/bin/perl # # Reply with proper protocol # print "Content-type: text/html\n\n"; # # Get the data from the form ... # $input = <STDIN>; # # ... and echo them back # print "$input\n\n\n"; $input =~ s/%..//g; $input =~ s/\+/ /g; @array = split('&',$input); foreach $var ( @array ) { print "$var<br>"; } print "<hr>\n"; ($name,$variable1) = split("variable1=",$array[0]); ($name,$variable2) = split("variable2=",$array[1]); print "<br>var1 = $variable1<br>"; print "<br>var2 = $variable2<br>"; print "<br>Done! \n";
and the output
variable1=Mark+Burgess&variable2=%0D%0AI+just+called+to+say (wrap) +....%0D%0A...hey+pig%2C+nothing%27s+working+out+the+way+I+planned variable1=Mark Burgess variable2=I just called to say .......hey pig nothings working (wrap) out the way I planned var1 = Mark Burgess var2 = I just called to say .......hey pig nothings working out (wrap) the way I planned Done!
Let us now use this technique to develop a guest book application. Based on the code above, analyze the following code.
#!/local/bin/perl #################################################################### # # Guest book # #################################################################### $guestbook_page = "/iu/nexus/ud/mark/www/tmp/cfguest.html"; $tmp_page = "/iu/nexus/ud/mark/www/tmp/guests.tmp"; $remote_host = $ENV{"REMOTE_HOST"}; print "Content-type: text/html\n\n"; print "<br><hr><br>\n"; print "Thank you for submitting your comment!<br><br>\n"; print "best wishes,<br><br>"; print "-Mark<br><br><br>"; print "Return to <a href=http://www.iu.hio.no/~mark/menu.html>menu</a>\n"; $input = <STDIN>; $input =~ s/%..//g; $input =~ s/\+/ /g; @array = split('&',$input); ($skip,$name) = split("var1=",$array[0]); ($skip,$message) = split("var2=",$array[1]); if (! open (PAGE, $guestbook_page)) { print "Content-type: text/html\n\n"; print "couldn't open guestbook page file!"; } if (! open (TMP, "+>$tmp_page")) { print "Content-type: text/html\n\n"; print "couldn't open temporary output file!"; } while ($line = <PAGE>) { if ($line =~ /<h3>Number of entries: (..)/) { $entry_no = $1; $entry_no++; $line = "<h3>Number of entries: $entry_no </h3>\n"; } if ($line =~ /<!-- LAST ENTRY -->/) { $date = `date +"%A, %b %d %Y"`; print TMP "<b>Entry $date from host: $remote_host</b>\n<p>\n"; print TMP "From: $name\n<p>\n"; print TMP $message; print TMP "\n<hr>\n"; } print TMP "$line"; } close PAGE; close TMP; if (! rename ($tmp_page, $guestbook_page)) { print "Oops! Rename operation failed!\n"; } chmod (0600, $guestbook_page);
This script works by reading through the old guest book file, opening a new copy of the guest book file and appending a new messages at the end. The end of the message section (not counting the `</HTML>' tags) is marked by a comment line.
<!-- LAST ENTRY -->
Note that a provisional guest book file has to exist in the first place. The script writes to a new file and then swaps the new file for the old one. The guest book file looks something like this:
<html><head> <title>Comments</title> </head> <body> <h1>My guest book</h1> <b>Entry no. Wednesday, Feb 28 1996 from host: dax</b> <p> From: Mark.Burgess@iu.hio.no <p> Just to start the ball rolling.... <hr> <b>Entry no. Tuesday, Mar 26 1996 from host: enterprise.subspace.net</b> <p> From: spock@enterprise <p> Registering a form of energy never before encountered. <!-- LAST ENTRY --> </body> <address><a href="http://www.iu.hio.no/~mark">Mark Burgess</a> - Mark.Burgess@iu.hio.no</addre ss> </html>
The directory in which this file lies needs to be writable to
the user nobody (the WWW user) and the files within need to be
deletable by nobody but no one else. Some users try to make
guest book scripts setuid-themselves in order to overcome the
problem that httpd runs with uid nobody, but this opens many
security issues. In short it is asking for trouble. Unfortunately
an ordinary user cannot use chown
in order to give access
only to the WWW user nobody, so this approach needs the cooperation
of the system administrator. Nevertheless this is the most secure
approach.
Try to work through this example step for step.
The PHP language makes the whole business of web programming rather simpler than perl. It hides the business of translating variables from forms into new variables in a CGI program and it even allows you to embed active code into you HTML pages. PHP has special support for querying data in an SQL database like MySQL or Oracle. PHP documentation lives at @uref{http://www.php.net}.
PHP code can be embedded inside HTML pages provided your WWW server is configurered with PHP support. PHP code lives inside a tag with the general form
<?php code... ?>
For example, we could use this to import one file into another and print out a table of numbers:
<html> <body> <?php include "file.html" for ($i = 0; $i < 10; $i++) { print "Counting $i<br>"; } ?> </body> </html>
This makes it easy to generate WWW pages with a fixed visual layout:
<?php # # Standard layout # # Set $title, $comment and $contents ########################################################################## print "<body>\n"; print "<img src=img/header.gif>"; print "<h1>"$title</h1>"; print "<em>$comment</em>"; print "<blockquote>\n"; include $contents; print ("</blockquote>\n"); print ("</body>\n"); print ("</html>\n");
Variables are easily set by calling PHP code in the form of a CGI program from a form.
PHP is particularly good at dealing with forms, as a CGI scripting language. Consider the following form:
<html> <body> <form action="/cgi-bin-scriptalias/spititout.php" method="post"> Name: <input type="text" name="personal[name]"><br> Email: <input type="text" name="personal[email]"><br> Preferred language: <select multiple name="language[]"> <option value="English">English <option value="Norwegian">Norwegian <option value="Gobbledigook">Gobbledigook </select> <input type=image src="image.gif" name="sub"> </form> </body> </html>
This produces a page into which one types a name and email address and chooses a language from a list of three possible choices. When the user clicks on a button marked by the file `image.gif' the form is posted. Here is a program which unravels the data sent to the CGI program:
#!/local/bin/php <?php # # A CGI program which handles a form # Variables are translated automatically # $title = "This page title"; $comment = "This pages talks about the following....."; ########################################################################## echo "<body>"; echo "<h1>$title</h1>"; echo "<em>$comment</em>"; echo "<blockquote>\n"; ### echo "Your name is $personal[name]<br><br>"; echo "Your email is $personal[email]<br><br>"; echo "Language options: "; echo "<table> "; for ($i = 0; strlen($language[$i]) > 0; $i++) { echo "<tr><td bgcolor=#ff0000>Variable language[$i] = $language[$i]</td></tr>"; } if ($language[0] == "Norwegian") { echo "Hei alle sammen<p>"; } else { echo "Greetings everyone, this page will be in English<p>"; } echo "</table> "; ### echo ("</blockquote>\n"); echo ("</body>\n"); echo ("</html>\n"); ?>
If your web-server supports PHP there is no need for separate
CGI-scripts handling form output. A single PHP-script can
create the form and handle the output simultaneously. In addition
this script can be placed wherever the web-server is able to read
HTML files. PHP defines a special variable $PHP_SELF
which
provides the action=
assignment of the form with the script
itself. Moreover such PHP-scripts checks whether the user has
submitted any data by checking if the variables of the form is set
with the command isset()
. The following code shows how easily
a guestbook can be made using PHP compared to the Perl-code shown
in a previous section. See section A complete guestbook example in perl.
<html> <head> <title>My Guestbook</title> </head> <body> <h1>Welcome to my Guestbook</h1> <h2>Please write me a little note below</h2> <form action="<?echo $PHP_SELF?>" method="POST"> <textarea cols=40 rows=5 name=note wrap=virtual></textarea> <input type=submit value=" Send it "> </form> <?php $file = "/iu/nexus/ud/haugerud/www/cgi-out/guestbook.txt"; if(isset($note)) { $date = date("F j, Y, G:i"); $buffer = "<h3>Message sent from IP-address $REMOTE_ADDR</h3>\n"; $buffer .= "<h4>$date</h4>\n"; $buffer .= nl2br($note).'<br>'; $handle = fopen ($file, "r"); while (!feof($handle)) { $buffer .= fread($handle,4096); } fclose($handle); $outhandle=fopen ($file,"w"); fputs($outhandle,$buffer); fclose($outhandle); } ?> <h2>The entries so far:</h2> <? @ReadFile($file) ?> </body> </html>
This section is not meant to teach you C. It is a guide to using C in UNIX and it is assumed that you have a working knowledge of the language. See the GNU C-Tutorial for an introduction to basics.
In the preceding chapters we have been looking at ways to get simple programming tasks done. The immediateness of the script languages is a great advantage when we just want to get a job done as quickly as possible. Scripts lend themselves to simple system administration tasks like file processing, but they do not easily lend themselves to more serious programs.
Although some system administrators have grown to the idea that shell programming is easier, I would argue that this is not really true. First of all, most of the UNIX shell commands are just wrapper programs for C function calls. Why use the wrapper when you can use the real thing? Secondly, the C function calls return data in pointers and structures which are very easy to manipulate, whereas piping the output of shell programs into others can be a very messy and awkward way of working. Here are some of the reasons why we also need a more traditional programming language like C.
A C program consists of a set of function, beginning with the main program:
main () /* This is a comment */ { Commands ... }
The source code of a C program can be divided into several text files.
C compiles all functions separately; the linker ld
joins them all
up at the end. This means that we can plan out a strategy for writing
large programs in a clear and efficient manner.
NOTE: C++ style comments `//...' are not allowed by most C compilers.
Most UNIX systems now have ANSI C compatible compilers, but this has not always been the case. Most UNIX programs written in a version of C which is older than the ANSI standard, so you will need an appreciation of old Kernighan and Ritchie C conventions for C programming. See for example my C book.
An obvious difference between ANSI C and K&R C is that the C++ additions to the language are not included. Here are some useful points to remember.
const int blah = 1;use
#define blah 1Remember that the hash symbol `#' must be the first character on a line under UNIX.
void function (char *string, int a, int b) { }Instead one writes:
void function (string, a, b) char *string; int a,b; { }
Most UNIX programs are very large and are split up into many files. Remember, when you split up programs into several files, you must declare variables as `extern' in file A if they are really declared in file B. in which you want to use them. This tells the compiler that it should not try to create local storage for the variable, because this was already done in another file.
Most of the system calls in UNIX return data in the form of `struct' variables. Sometimes these are structures used by the operating system itself -- in other cases they are just put together so that programmers can handle a packet of data in a convenient way.
If in doubt, you can find the definitions of these structures in the relevant include files under `/usr/include'.
Since UNIX comes in many flavours the system calls are not always compatible and may have different options and arguments. Because of this there is a number of standardizing organizations for UNIX. One of them is POSIX which is an organization run by the major UNIX vendors. Programs written for UNIX are now expected to be POSIX compliant. This is not something you need to think about at the level of this course, but you should certainly remember that there exist programming standards and that these should be adhered to. The aim is to work towards a single standard UNIX.
The C compiler on the UNIX system is traditionally called `cc' and has always been a traditional part of every UNIX environment. Recently several UNIX vendors have stopped including the C compiler as a part of their operating systems and instead sell a compiler separately. Fortunately there is a public domain Free Software version of the compiler called `gcc' (the GNU C compiler). We shall use this in all the examples.
To compile a program consisting of several files of code, we first compile all of the separate pieces without trying to link them. There are therefore two stages: first we turn `.c' files into `.o' files. This compiles code but does not fix any address references. Then we link all `.o' files into the final executable, including any libraries which are used.
Let's suppose we have files `a.c', `b.c' and `c.c'. We write:
gcc -c a.c b.c c.c
This creates files `a.o', `b.o' and `c.o'. Next we link them into one file called `myprog'.
gcc -o myprog a.o b.o c.o
If the naming option `-o myprog' is not used, the link `ld' uses the default name a.out for the executable file.
The resulting file is called `myprog' and includes references only to the standard library `libc'. If we wish to link in the math library `libm' or the cursor movement library `libcurses' -- or in general, a library called `libBLAH' , we need to use the `-l' directive.
gcc -o myprog files.o -lm -lcurses -lBLAH
The compiler looks for a suitable library in all of the directories listed in the environment variable `LD_LIBRARY_PATH'. Alternatively we can add a directory to the search path by using the `-L'. option:
gcc -o myprog files.o -L/usr/local/lib -lm -lcurses -lBLAH
Normally the compiler looks for include files only in the directory `/usr/include'. We can add further paths to search using the `-I' option.
gcc -o myprog file.c -I/usr/local/include -I/usr/local/X11/include
Previously, UNIX libraries have been in `a.out' code format, but recent releases of UNIX have gone over to a more efficient and flexible format called ELF (executable and linking format).
Libraries are collections of C functions which the operating system
creators have written for our convenience. The source code for such a
library is just the source for a collection of functions -- there is no
main
program.
There are two kinds of library used by modern operating systems:
archive libraries or static libraries
and shared libraries
or dynamical libraries. An archive
library has a name of the form
libname.a
When an archive library is linked to a program, it is appended lock, stock and barrel to the program code. This uses a lot of disk space and makes the size of the compiled program very large. Shared libraries (shared objects `so' or shared archives `sa' generally have names of the form)
libname.so libname.sa
often with version numbers appended. When a program is linked with a shared library the code is not appended to the program. Instead pointers to the shared objects are created and the library is loaded at runtime, thus avoiding the problem of having to store the library effectively multiple times on the disk.
To make an archive library we compile all of the functions we wish to include in the library
gcc -c function1.c function2.c ...
and then join the files using the `ar' command.
ar rcv libMYLIB.a function1.o ar rcv libMYLIB.a function2.o
To make a shared library one provides an option to the linker
program. The exact method is different in different operating systems,
so you should look at the manual page for ld
on your system.
Under SunOS 4 we take the object files `*.o' and run
ld -o libMYLIB.so.1.1 -assert pure-text *.o
Under HPUX, we write
ld -b -o libMYLIB.so.1.1 *.o
With the GNU linker, you write
ld -shared -o libMYLIB.so.1.1 *.o
NOTE: when you add a shared library to the system under SunOS or GNU/Linux you must run the command `ldconfig', making sure that the path to the library is included in `LD_LIBRARY_PATH'. SunOS and GNU/Linux use a cache file `/etc/ld.so.cache' to keep current versions of libraries. GNU/Linux also uses a configuration file called `/etc/ld.so.conf'.
It is important to understand how the C compiler finds the files it needs. We have already mentioned the `-I' and `-L' options to the compilation command line. In general, all system include files can be found in the directory `/usr/include' and subdirectories of this directory. All system libraries can be found in `/usr/lib'.
Many packages build their own libraries and keep the relevant files
in separate directories so that if the system gets reinstalled, they
do not get deleted. This is true for example of the X-windows system.
The include and library files for this are typically kept in
directories which look something like `/usr/local/X11R5/include'
and `/usr/X11R6/lib'. That means that we need to give all of
this information to the compiler. Compiling a program becomes
a complicated task in many cases so we need some kind of script
to help us perform the task. The UNIX tool make
was designed
for this purpose.
Nowadays compilers are often sold with fancy user environments driven by menus which make it easier to compile programs. UNIX has similar environments but all of them use shell-based command line compilation beneath the surface. That is because UNIX programmers are used to writing large and complex programs which occupy many directories and subdirectories. Each directory has to be adapted or configured to fit the particular flavour of UNIX system it is being compiled upon. Interactive user environments are very poor at performing this kind of service. UNIX solves the problem of compiling enormous trees of software (such as the UNIX system itself!) by using a compilation language called `make'. Such language files can be generated automatically by scripts, allowing very complex programs to configure and compile themselves from a single control script.
Typing lines like
cc -c file1.c file2.c ... cc -o target file1.o ....
repeatedly to compile a complicated program can be a real nuisance. One possibility would therefore be to keep all the commands in a script. This could waste a lot of time though. Suppose you are working on a big project which consists of many lines of source code -- but are editing only one file. You really only want to recompile the file you are working on and then relink the resulting object file with all of the other object files. Recompiling the other files which hadn't changed would be a waste of time. But that would mean that you would have to change the script each time you change what you need to compile.
A better solution is to use the `make' command. `make' was designed for precisely this purpose. To use `make', we create a file called `Makefile' in the same directory as our program. `make' is a quite general program for building software. It is not specifically tied to the C programming language--- it can be used in any programming language.
A `make' configuration file, called a `Makefile', contains
rules which describe how to compile or build all of the pieces of
a program. For example, even without telling it specifically,
make
knows that in order to go from `prog.c' to
`prog.o' the command `cc -c prog.c' must be executed. A
Makefile works by making such associations. The Makefile contains a
list of all of the files which compose the program and rules as to how
to get to the finished product from the source.
The idea is that, to compile a program, we just have to type make. `make' then reads the Makefile and compiles all of the parts which need compiling. It does not recompile files which have not changed since the last compilation! How does it do this? `make' works by comparing the time-stamp on the file it needs to create with the time-stamp on the file which is to be compiled. If the compiled version exists and is newer than its source then the source does not need to be recompiled.
To make this idea work in practice, `make' has to know how to go through the steps of compiling a program. Some default rules are defined in a global configuration file, e.g.
/usr/include/make/default.mk
Let's consider an example of what happens for the the three files `a.c', `b.c' and `c.c' in the example above -- and let's not worry about what the Makefile looks like yet.
The first time we compile, only the `.c' files exist. When we type `make', the program looks at its rules and finds that it has to make a file called `myprog'. To make this it needs to execute the command
gcc -o myprog a.o b.o c.o
So it looks for `a.o' etc and doesn't find them. It now goes to a kind of subroutine and looks to see if it has any rules for making files called `.o' and it discovers that these are made by compiling with the `gcc -c' option. Since the files do not exist, it does this. Now the files `a.o b.o c.o' exist and it jumps back to the original problem of trying to make `myprog'. All the files it needs now exist and so it executes the command and builds `myprog'.
If we now edit `a.c', and type `make' once again -- it goes through the same procedure as before but now it finds all of the files. So it compares the dates on the files -- if the source is newer than the result, it recompiles.
By using this recursive method, `make' only compiles those parts of a program which need compiling.
To write a Makefile, we have to tell `make' about dependencies. The dependencies of a file are all of those files which are required to build it. Thus, the dependencies of `myprog' are `a.o', `b.o' and `c.o'. The dependencies of `a.o' are simply `a.c', the dependencies of `b.o' are `b.c' and so on.
A Makefile consists of rules of the form:
target : dependencies TAB rule;
The target is the thing we want to build, the dependencies are like subroutines to be executed first if they do not exist. Finally the rule is to be executed if all if the dependencies exist; it takes the dependencies and turns them into the target. There are two important things to remember:
Let's look at an example Makefile for a program which consists of two
course files `main.c' and `other.c' and which makes use of a
library called `libdb' which lies in the directory `/usr/local/lib'.
Our aim is to build a program called database
:
# # Simple Makefile for `database' # # First define a macro OBJ = main.o other.o CC = gcc CFLAGS = -I/usr/local/include LDFLAGS = -L/usr/local/lib -ldb INSTALLDIR = /usr/local/bin # # Rules start here. Note that the $@ variable becomes the name of the # executable file. In this case it is taken from the ${OBJ} variable # database: ${OBJ} ${CC} -o $@ ${OBJ} ${LDFLAGS} # # If a header file changes, normally we need to recompile everything. # There is no way that make can know this unless we write a rule which # forces it to rebuild all .o files if the header file changes... # ${OBJ}: ${HEADERS} # # As well as special rules for special files we can also define a # "suffix rule". This is a rule which tells us how to build all files # of a certain type. Here is a rule to get .o files from .c files. # The $< variable is like $? but is only used in suffix rules. # .c.o: ${CC} -c ${CFLAGS} $< ####################################################################### # Clean up ####################################################################### # # Make can also perform ordinary shell command jobs # "make tidy" here performs a cleanup operation # clean: rm -f ${OBJ} rm -f y.tab.c lex.yy.c y.tab.h rm -f y.tab lex.yy rm -f *% *~ *.o rm -f mconfig.tab.c mconfig.tab.h a.out rm -f man.dvi man.aux man.log man.toc rm -f cfengine.tar.gz cfengine.tar cfengine.tar.Z make tidy rm -f cfengine install: ${INSTALLDIR}/database cp database ${INSTALLDIR}/database
The Makefile above can be invoked in several ways.
make make database make clean make install
If we simple type `make' i.e. the first of these choices, `make' takes the first of the rules it finds as the object to build. In this case the rule is `database', so the first two forms above are equivalent.
On the other hand, if we type
make clean
then execution starts at the rule for `clean', which is normally used to remove all files except the original source code. Make `install' causes the compiled program to be installed at its intended destination.
`make' uses some special variables (which resemble the special variables used in Perl -- but don't confuse them). The most useful one is `$@' which represents the current target -- or the object which `make' would like to compile. i.e. as `make' checks each file it would like to compile, `$@' is set to the current filename.
$@
$?
target: file1.o file2.o TAB cc -o $@ $?
$<
Note that, because `make' has some default rules defined in its configuration file, a single-file C program can be compiled very easily by typing
make filename.c
This is equivalent to
cc -c filename.c cc -o filename filename.o
Standard rules for C++ are not often built into UNIX systems at the time of writing, but we can create them in our own Makefiles very easily. Here we shall use the GNU compiler g++'s conventions for C++ files. Here is a sample Makefile for using C++. Note that the `.SUFFIXES' command must be used to declare new endings or file extensions.
################################################################## # # This is the Makefile for g++ # ################################################################## OBJ = cpp-prog.o X.o Y.o Z.o CCPLUS = g++ .SUFFIXES: .C .o .h # # Program Rules # filesys: ${OBJ} $(CCPLUS) -o filesys $(OBJ) # # Extra dependencies on the header file # (if the header file changes, we need to rebuild *.o) # cpp-prog.o: filesys.h X.o: filesys.h Y.o: filesys.h Z.o: filesys.h # # Suffix rules # .C.o: $(CCPLUS) -c $<
The general rule here tells make
that a `.o' file
can be created from a `.C' file by executing the
command `$(CCPLUS) -c'. (This is identical to the
C case, except for the name of the compiler). The extra
dependencies tell make
that, if we change the header
file `filesys.h', then we must recompile all the files
which read in `filesys.h', since this could affect all of
these. Finally, the highest level rule says that to make
`filesys' from the `.o' files, we have to run
`$(CCPLUS) -o filesys *.o'.
argv
, argc
and envp
parametersWhen we write C programs which reads command line arguments, they are fed to us as an array of strings called the argument vector. The mechanisms for the C-shell and Perl are derived from the C argument vector. To read in the command line, we write
main (argc,argv,envp) int argc; char *argv[], *envp[]; { printf ("The first argument was %s\n",argv[1]); }
Argument zero is the name of the program itself and `argv[argc-1]' is the last argument. The above definitions are in Kernighan and Ritchie C style. In ANSI C, the arguments can be declared using prototype:
main (int argc, char **argv) { }
The array of strings `envp[]' is a list of values of the environment variables of the system, formatted by
NAME=value
This gives C programmers access to the shell's global environment.
In addition to the `envp' vector, it is possible to access the environment variables through the call `getenv()'. This is used as follows; suppose we want to access the shell environment variable `$HOME'.
char *string; string = getenv("HOME");
`string' is now a pointer to static but public data. You should not use `string' as if it were you're own property because it will be used again by the system. Copy it's contents to another string before using the data.
char buffer[500]; strcpy (buffer,string);
All of the regular C functions from the standard library are available to UNIX programmers. The standard functions only address the issue of reading and writing to files however, they do not deal with operating system specific attributes such as file permissions and file types. Nor is there a mechanisms for obtaining lists of files within a directory. The reason for these omissions is that they are operating system dependent. To find out about these other attributes POSIX describes some standard UNIX system calls.
Files and directories are handled by functions defined in the header file `dirent.h'. In earlier UNIX systems the file `dir.h' was used -- and the definitions were slightly different, but not much. To get a list of files in a directory we must open the directory and read from it -- just like a file. (A directory is just a file which contains data on its entries). The commands are
opendir closedir readdir
See the manual pages for dirent. These functions return pointers to a
dirent
structure which is defined in the file
`/usr/include/dirent.h'. Here is an example ls
command
which lists the contents of the directory `/etc'. This header
defines a structure
struct dirent { off_t d_off; /* offset of next disk dir entry */ unsigned long d_fileno; /* file number of entry */ unsigned short d_reclen; /* length of this record */ unsigned short d_namlen; /* length of string in d_name */ char d_name[255+1]; /* name (up to MAXNAMLEN + 1) */ };
which can be used to obtain information from the directory nodes.
#include <stdio.h> #include <dirent.h> main () { DIR *dirh; struct dirent *dirp; static char mydir[20] = "/etc"; if ((dirh = opendir(mydir)) == NULL) { perror("opendir"); return; } for (dirp = readdir(dirh); dirp != NULL; dirp = readdir(dirh)) { printf("Got dir entry: %s\n",dirp->d_name); } closedir(dirh); }
Notice that reading from a directory is like reading from a file
with fgets()
,
but the entries are filenames rather than lines of text.
stat()
To determine the file properties or statistics we use the function call `stat()' or its corollary `lstat()'. Both these functions find out information about files (permissions, owner, filetype etc). The only difference between them is the way in which they treat symbolic links. If `stat' is used on a symbolic link, it stats the file the link points to rather than the link itself. If `lstat' is used, the data refer to the link. Thus, to detect a link, we must use `lstat', See section lstat and readlink.
The data in the `stat' structure are defined in the file `/usr/include/sys/stat.h'. Here are the most important structures.
struct stat { dev_t st_dev; /* device number*/ ino_t st_ino; /* file inode */ mode_t st_mode; /* permission */ short st_nlink; /* Number of hardlinks to file */ uid_t st_uid; /* user id */ gid_t st_gid; /* group id */ dev_t st_rdev; off_t st_size; /* size in bytes */ time_t st_atime; /* time file last accessed */ time_t st_mtime; /* time file contents last modified */ time_t st_ctime; /* time last attribute change */ long st_blksize; long st_blocks; };
The function `stat()' treats symbolic links as though they were the files they point to. In other words, if we use `stat()' to read a symbolic link, we end up reading the file the link points to and not the link itself--- we never see symbolic links. To avoid this problem, there is a different version of the stat function called `lstat()' which is identical to `stat()' except that it treats links as links and not as the files they point to. This means that we can test whether a file is a symbolic link, only if we use `lstat()'. (See the next paragraph.)
Once we have identified a file to be a symbolic link, we use the `readlink()' function to obtain the name of the file the link points to.
#define bufsize 512 char buffer[bufsize]; readlink("/path/to/file",buffer,bufsize);
The result is returned in the string buffer.
stat()
test macrosAs we have already mentioned, the UNIX mode bits contain not only information about what permissions a file has, but also bits describing the type of file -- whether it is a directory or a link etc. There are macros defined in UNIX to extract this information from the `st_mode' member of the `stat' structure. They are defined in the `stat.h' headerfile. Here are some examples.
#define S_ISBLK(m) /* is block device */ #define S_ISCHR(m) /* is character device */ #define S_ISDIR(m) /* is directory */ #define S_ISFIFO(m) /* is fifo pipe/socket */ #define S_ISREG(m) /* is regular (normal) file */ #define S_ISLNK(m) /* is symbolic link */ /* Not POSIX */ #define S_ISSOCK(m) /* is a lock */ #define S_IRWXU /* rwx, owner */ #define S_IRUSR /* read permission, owner */ #define S_IWUSR /* write permission, owner */ #define S_IXUSR /* execute/search permission, owner */ #define S_IRWXG /* rwx, group */ #define S_IRGRP /* read permission, group */ #define S_IWGRP /* write permission, grougroup */ #define S_IXGRP /* execute/search permission, group */ #define S_IRWXO /* rwx, other */ #define S_IROTH /* read permission, other */ #define S_IWOTH /* write permission, other */ #define S_IXOTH /* execute/search permission, other */
These return true or false when acting on the mode member. Here is an example See section Example filing program.
struct stat statvar; stat("file",&statvar); /* test return values */ if (S_ISDIR(statvar.st_mode)) { printf("Is a directory!"); }
The following example program demonstrates the use of the directory
functions in dirent
and the stat
function call.
/********************************************************************/ /* */ /* Reading directories and `statting' files */ /* */ /********************************************************************/ #include <stdio.h> #include <dirent.h> #include <sys/types.h> #include <sys/stat.h> #define DIRNAME "/." #define bufsize 255 /********************************************************************/ main () { DIR *dirh; struct dirent *dirp; struct stat statbuf; char *pathname[bufsize]; char *linkname[bufsize]; if ((dirh = opendir(DIRNAME)) == NULL) { perror("opendir"); exit(1); } for (dirp = readdir(dirh); dirp != NULL; dirp = readdir(dirh)) { if (strcmp(".",dirp->d_name) == 0 || strcmp("..",dirp->d_name) == 0) { continue; } if (strcmp("lost+found",dirp->d_name) == 0) { continue; } sprintf(pathname,"%s/%s",DIRNAME,dirp->d_name); if (lstat(pathname,&statbuf) == -1) /* see man stat */ { perror("stat"); continue; } if (S_ISREG(statbuf.st_mode)) { printf("%s is a regular file\n",pathname); }; if (S_ISDIR(statbuf.st_mode)) { printf("%s is a directory\n",pathname); } if (S_ISLNK(statbuf.st_mode)) { bzero(linkname,bufsize); /* clear string */ readlink(pathname,linkname,bufsize); printf("%s is a link to %s\n",pathname,linkname); } printf("The mode of %s is %o\n\n",pathname,statbuf.st_mode & 07777); } closedir(dirh); }
fork()
, exec()
, popen()
and systemThere is a number of ways in which processes can interact with one another and in which we can control their behaviour. We shall not go into great detail in this course, only provide examples for reference.
The UNIX `fork()' function is used to create child processes. This is the basis of all `heavyweight' multitasking under UNIX. Here is a simple example of fork in which we start a child process from within a program and wait for it to finish. Note that the code for the parent and the child is is the same file. The only thing that distinguishes parent from child is the value returned by the fork function.
When `fork()' is called, it duplicates the entire current process so that two parallel processes are then running. The only difference between these is that the child process (the copy) gets a return value of zero from `fork()', whereas the parent gets a return value equal to the process identifier (pid) of the child. This value can be used by the parent to send messages or to wait for the child. Here we show a simple example in which the `wait(NULL)' command is used to wait for the last child spawned by the parent.
/**************************************************************/ /* */ /* A brief demo of the UNIX process duplicator fork(). */ /* */ /**************************************************************/ #include <stdio.h> /***************************************************************/ main () { int pid, cid; pid = getpid(); printf ("Fork demo! I am the parent (pid = %d)\n",pid); if (! fork()) { cid = getpid(); printf ("I am the child (cid = %d) of (pid=%d)\n",cid,pid); ChildProcess(); exit(0); } printf("Parent waiting here for the child...\n"); wait(NULL); printf("Child finished, parent quitting too!\n"); } /**************************************************************/ ChildProcess() { int i; for (i = 0; i < 10; i++) { printf ("%d...\n",i); sleep(1); } }
Another possibility is that we might want to execute a program and
wait to find out what the result of the program is before
continuing. There are two ways to do this. The first is a variation
on the theme above and uses fork()
.
Let's create a function which runs a shell command from within a C program, and determines its return value. We make the result a boolean (integer) value, so that the function returns `true' if the shell command exits normally, See section Return codes.
if (ShellCommandReturnsZero(shell-command)) { printf ("Command %s went ok\n",shell-command); }
To do this we first have to fork a new process and then
use one of the exec
commands to load a new
code image on top of the new process.
shell commands from C This sounds complicated, but it is necessary
because of the way UNIX handles processes. If we had no use for the
return value, we could simply execute a shell command using the
system("shell command")
function, (which does all this for us)
but when system()
exits, we can only tell if the command was
executed successfully or unsuccessfully--we learn nothing about what
actually failed (the shell or command which was executed under the
shell?) If we require detailed information about what happened to the
child process then we need to do the following.
#include <sys/types.h> #include <sys/wait.h> /* Send complete command as a string */ /* including all arguments */ ShellCommandReturnsZero(comm) char *comm; { int status, i, argc; pid_t pid; char arg[maxshellargs][bufsize]; char **argv; /* Build argument array for execv call*/ for (i = 0; i < maxshellargs; i++) { bzero (arg[i],bufsize); } argc = SplitCommand(comm,arg); if ((pid = fork()) < 0) { FatalError("Failed to fork new process"); } else if (pid == 0) /* child */ { argv = malloc((argc+1)*sizeof(char *)); for (i = 0; i < argc; i++) { argv[i] = arg[i]; } argv[i] = (char *) NULL; if (execv(arg[0],argv) == -1) { yyerror("script failed"); perror("execvp"); exit(1); } } else /* parent */ { if (wait(&status) != pid) { printf("Wait for child failed\n"); perror("wait"); return false; } else { if (WIFSIGNALED(status)) { printf("Script %s returned: %s\n",comm,WTERMSIG(status)); return false; } if (! WIFEXITED(status)) { return false; } if (WEXITSTATUS(status) == 0) { return true; } else { return false; } } } } /*******************************************************************/ SplitCommand(comm,arg) char *comm, arg[maxshellargs][bufsize]; { char *sp; int i = 0, j; char buff[bufsize]; for (sp = comm; *sp != NULL; sp++) { bzero(buff,bufsize); if (i >= maxshellargs-1) { yyerror("Too many arguments in embedded script"); FatalError("Use a wrapper"); } while (*sp == ' ' || *sp == '\t') { sp++; } switch (*sp) { case '\"': sscanf (++sp,"%[^\"]",buff); break; case '\": sscanf (++sp,"%[^\']",buff); break; default: sscanf (sp,"%s",buff); break; } for (j = 0; j < bufsize; j++) { arg[i][j] = buff[j]; } sp += strlen(arg[i]); i++; } return (i); }
In this example, the script waits for the exit signal from the child
process before continuing. The return value from the child is
available from the wait function with the help of a set of macros
defined in `/usr/include/sys/wait.h'. The value is
given by WTERMSIG(status)
.
In the final example, we can open a pipe to a
process directly in a C program
as though it were a file, by using the function popen()
.
Pipes may be opened for reading or for writing, in exactly the same
way as a file is opened. The child process is automatically
synchronized with the parent using this method. Here is a
program which opens a UNIX command for reading (both stdout and
stderr) from the child process are piped into the program.
Notice that the syntax used in this call is that used by
the Bourne shell, since this is build deeply into the UNIX
execution design.
#define bufsize 1024 FILE *pp; char VBUFF[bufsize]; ... if ((pp = popen( "/sbin/mount -va 2<&1","r")) == NULL) { printf("Failed to open pipe\n"); return errorcode; } while (!feof(pp)) { fgets(VBUFF,bufsize,pp); /* Just write the output to stdout */ printf ("Pipe read: %s\n",VBUFF); } pclose(pp);
popen()
One problem with the popen()
system call is that it uses a
shell to execute the command it obtains a pipe to. In the past this
has been used to allow UNIX security breaches, using a so-called
IFS
attack which can trick the shell into executing a program with
the name of the first node in the directory of the executable. For instance,if the pipe was to open the program `/bin/ps', this coudl be tricked
into executing a program in the current working directory of the process
called `bin' with argument `ps'.
The solution is not to use a shell at all, but to replace popen()
with a version which calls exec()
directly. Here is a safe
version from the source code of cfengine:
#define bufsize 4096 #define maxshellargs 20 pid_t *CHILD; int MAXFD = 20; /* Max number of simultaneous pipes */ /***************************************************************/ FILE *cfpopen(command, type) char *command, *type; { char arg[maxshellargs][bufsize]; int i, argc, pd[2]; char **argv; pid_t pid; FILE *pp = NULL; if ((*type != 'r' && *type != 'w') || (type[1] != '\0')) { errno = EINVAL; return NULL; } if (CHILD == NULL) /* first time */ { if ((CHILD = calloc(MAXFD,sizeof(pid_t))) == NULL) { return NULL; } } if (pipe(pd) < 0) /* Create a pair of descriptors to this process */ { return NULL; } if ((pid = fork()) == -1) { return NULL; } if (pid == 0) { switch (*type) { case 'r': close(pd[0]); /* Don't need output from parent */ if (pd[1] != 1) { dup2(pd[1],1); /* Attach pp=pd[1] to our stdout */ dup2(pd[1],2); /* Merge stdout/stderr */ close(pd[1]); } break; case 'w': close(pd[1]); if (pd[0] != 0) { dup2(pd[0],0); close(pd[0]); } } for (i = 0; i < MAXFD; i++) { if (CHILD[i] > 0) { close(CHILD[i]); } argc = SplitCommand(command,arg); argv = (char **) malloc((argc+1)*sizeof(char *)); if (argv == NULL) { FatalError("Out of memory"); } for (i = 0; i < argc; i++) { argv[i] = arg[i]; } argv[i] = (char *) NULL; if (execv(arg[0],argv) == -1) { sprintf(OUTPUT,"Couldn't run %s",arg[0]); CfLog(cferror,OUTPUT,"execv"); } _exit(1); } } else { switch (*type) { case 'r': close(pd[1]); if ((pp = fdopen(pd[0],type)) == NULL) { return NULL; } break; case 'w': close(pd[0]); if ((pp = fdopen(pd[1],type)) == NULL) { return NULL; } } CHILD[fileno(pp)] = pid; return pp; } } /***************************************************************/ cfpclose(pp) FILE *pp; { int fd, status; pid_t pid; Debug("cfpclose(pp)\n"); if (CHILD == NULL) /* popen hasn't been called */ { return -1; } fd = fileno(pp); if ((pid = CHILD[fd]) == 0) { return -1; } CHILD[fd] = 0; if (fclose(pp) == EOF) { return -1; } Debug("cfpopen - Waiting for process %d\n",pid); #ifdef HAVE_WAITPID while(waitpid(pid,&status,0) < 0) { if (errno != EINTR) { return -1; } } return status; #else if (wait(&status) != pid) { return -1; } else { if (WIFSIGNALED(status)) { return -1; } if (! WIFEXITED(status)) { return -1; } return (WEXITSTATUS(status)); } #endif } /*******************************************************************/ /* Command exec aids */ /*******************************************************************/ SplitCommand(comm,arg) char *comm, arg[maxshellargs][bufsize]; { char *sp; int i = 0, j; char buff[bufsize]; for (sp = comm; sp < comm+strlen(comm); sp++) { bzero(buff,bufsize); if (i >= maxshellargs-1) { CfLog(cferror,"Too many arguments in embedded script",""); FatalError("Use a wrapper"); } while (*sp == ' ' || *sp == '\t') { sp++; } switch (*sp) { case '\0': return(i-1); case '\"': sscanf (++sp,"%[^\"]",arg[i]); break; case '\": sscanf (++sp,"%[^\']",arg[i]); break; case '`': sscanf (++sp,"%[^`]",arg[i]); break; default: sscanf (sp,"%s",arg[i]); break; } sp += strlen(arg[i]); i++; } return (i); }
Processes can receive signals from the UNIX kernel at any time. Some of these signals terminate the execution of the program. This can cause problems if the program is in the middle of critical activity such as writing to a file. For that reason we can trap signals and provide our own routine for handling them in a special way.
A signal handler is made by calling the function `signal()' for each signal and by specifying a pointer to a function which will be called in the event of a signal. For example:
main () { int HandleSignal(); signal(SIGTERM,HandleSignal); } HandleSignal() { /* Tidy up and exit cleanly */ exit(0); }
`SIGTERM' is the usual signal sent by the command `kill'. There are many other signals which can be sent to programs. Here is list. You have to decide for yourself whether or not you want to provide your own signal handling function. To ignore a signal, you write
signal(SIGtype,SIG_IGN);
To remove a signal handler and re-activate a signal, you write
signal(SIGtype,SIG_DFL);
A regular expression is a pattern for matching strings of text. We have met regular expressions earlier in connection with the shell and Perl. Naturally these earlier encounters have their roots in C functions for handling expressions. A regular expression is used by first `compiling' it into a convenient data structure. Then a matching function is used to compare the expression with a test string. In this example program we show how a regular expression typed in as an argument to the program is found within strings of input entered on the keyboard.
#include <stdio.h> #include <regex.h> main (argc,argv) int argc; char **argv; { char buffer[1024]; regex_t rx; regmatch_t match; size_t nmatch = 1; if (regcomp(&rx, argv[1], REG_EXTENDED) != 0) { perror("regcomp"); return; } while (!feof(stdin)) { fgets(buffer,1024,stdin); if (regexec(&rx,buffer,1,&match,0) == 0) { printf("Matched:(%s) at %d to %d",buffer,match.rm_so,match.rm_eo); } } regfree(&rx); }
Here is an example of its use. The output of the program is in italics
% a.out xyz this is a string another string an xyz string Matched: (an xyz string ) at 3 to 6 another xyz zyxxyz string Matched: (another xyz xyz string ) at 8 to 11 % a.out 'xyz|abc' This is a string An abc string Matched: (An abc string ) at 3 to 6 Or an xyz string Matched: (Or an xyz string ) at 6 to 9
If you don't want the match data set &pm
to NULL
. To get
an exact match rather than a substring check that the bounds are 0 and
strlen(argv[1])-1
.
Encryption with the SSLeay library, compile with command
gcc crypto.c -I/usr/local/ssl/include -L/usr/local/ssl/lib -lcrypto
Example of normal triple DES encryption which works only on an 8-byte buffer:
/*****************************************************************************/ /* */ /* File: crypto.c */ /* */ /* Compile with: gcc program.c -lcrypto (SSLeay) */ /* */ /*****************************************************************************/ #include <stdio.h> #include <des.h> #define bufsize 1024 /* Note how this truncates to 8 characters */ main () { char in[bufsize],out[bufsize],back[bufsize]; des_cblock key1,key2,key3,seed = {0xFE,0xDC,0xBA,0x98,0x76,0x54,0x32,0x10}; des_key_schedule ks1,ks2,ks3; strcpy(in,"1 2 3 4 5 6 7 8 9 a b c d e f g h i j k"); des_random_seed(seed); des_random_key(key1); des_random_key(key2); des_random_key(key3); des_set_key((C_Block *)key1,ks1); des_set_key((C_Block *)key2,ks2); des_set_key((C_Block *)key3,ks3); des_ecb3_encrypt((C_Block *)in,(C_Block *)out,ks1,ks2,ks3,DES_ENCRYPT); printf("Encrypted [%s] into [%s]\n",in,out); des_ecb3_encrypt((C_Block *)out,(C_Block *)back,ks1,ks2,ks3,DES_DECRYPT); printf("and back to.. [%s]\n",back); }
Triple DES, chaining mode, for longer strings (which must be a multiple of 8 bytes):
/*****************************************************************************/ /* */ /* File: crypto.c */ /* */ /* Compile with: gcc program.c -lcrypto (SSLeay) */ /* */ /*****************************************************************************/ #include <stdio.h> #include <des.h> #define bufsize 1024 /* This can be used on arbitrary length buffers */ main () { char in[bufsize],out[bufsize],back[bufsize],workvec[bufsize]; des_cblock key1,key2,key3,seed = {0xFE,0xDC,0xBA,0x98,0x76,0x54,0x32,0x10}; des_key_schedule ks1,ks2,ks3; strcpy(in,"1 2 3 4 5 6 7 8 9 a b c d e f g h i j k l m n o p q r s t u v w x y z"); des_random_seed(seed); des_random_key(key1); des_random_key(key2); des_random_key(key3); des_set_key((C_Block *)key1,ks1); des_set_key((C_Block *)key2,ks2); des_set_key((C_Block *)key3,ks3); /* This work vector can be intialized t anything ...*/ memset(workvec,0,bufsize); des_ede3_cbc_encrypt((C_Block *)in,(C_Block *)out,(long)strlen(in), ks1,ks2,ks3,(C_Block *)workvec,DES_ENCRYPT); printf("Encypted [%s] into [something]\n",in); /* .. but this must be initialized the same as above */ memset(workvec,0,bufsize); /* Note that the length is the original length, not strlen(out) */ des_ede3_cbc_encrypt((C_Block *)out,(C_Block *)back,(long)strlen(in), ks1,ks2,ks3,(C_Block *)workvec,DES_DECRYPT); printf("and back to.. [%s]\n",back); }
The C function `ioctl' (I/O control) is used to send special control commands to devices like the disk and the network interface. The syntax of the function is
int ioctl(fd, request, arg) int fd, request; long arg;
The first parameter is normally as device handle or socket descriptor. The second is a control parameter. Lists of valid control parameters are normally defined in the system `include' files for a particular device. They are device and system dependent so you need a local manual and som detective work to find out what they are. The final parameter is a pointer to a variable which receives return data from the device.
`ioctl' commands are device specific, by their nature. The commands for the ethernet interface device are only partially standardized, for example. We could read the ethernet device (which is called `le0' on a Sun workstation), using the following command:
# include <sys/socket.h> /* Typical includes for internet */ # include <sys/ioctl.h> # include <net/if.h> # include <netinet/in.h> # include <arpa/inet.h> # include <netdb.h> # include <sys/protosw.h> # include <net/route.h> struct ifreq IFR; int sk; struct sockaddr_in sin; strcpy(IFR.ifr_name,"le0"); IFR.ifr_addr.sa_family = AF_INET; if ((sk = socket(AF_INET,SOCK_DGRAM,IPPROTO_IP)) == -1) { perror("socket"); exit(1); } if (ioctl(sk,SIOCGIFFLAGS, (caddr_t) &IFR) == -1) { perror ("ioctl"); exit(1); }
We shall not go into the further details of `ioctl', but simply note its role in system programming.
DBT key,value; DB *dbp; DBC *dbcp; db_recno_t recno; if ((errno = db_open(CHECKSUMDB,DB_BTREE, DB_CREATE, 0664, NULL, NULL, &dbp)) != 0) { sprintf(OUTPUT,"cfd: couldn't open checksum database %s\n",CHECKSUMDB); CfLog(cferror,OUTPUT,"db_open"); return false; } bzero(&value,sizeof(value)); bzero(&key,sizeof(key)); key.data = filename; key.size = strlen(filename)+1; value.data = dbvalue; value.size = sizeof(dbvalue); if ((errno = dbp->del(dbp,NULL,&key,0)) != 0) { CfLog(cferror,"","db_store"); } key.data = filename; key.size = strlen(filename)+1; if ((errno = dbp->put(dbp,NULL,&key,&value,0)) != 0) { CfLog(cferror,"put failed","db->put"); } if ((errno = dbp->get(dbp,NULL,&key,&value,0)) == 0) { /* Not found ... */ return; } dbp->close(dbp,0);
This section is a taster only. You only need to know what lex and yacc are, not how they work.
`lex' and `yacc' are two tools for the C programmer who wishes to make a text parser. A text parser is a program which reads a text file and interprets the symbols in it. Every programming language must include a text parser, for instance.
The `yacc' (yet another compiler compiler) program generates C code which parses a textfile, given a description of the syntax rules for the file. In other words, we define the logical structure of the text file, according to the way we wish to interpret it and give the rules to `yacc'. `yacc' produces C code from this which does the job.
`lex' is a `lexer'. It is normally used together with `yacc'. `lex' tokenizes or identifies symbols in a file. What that means is that it reads in a file and matches types of string in the file which are defined in terms of regular expressions by the programmer, and returns symbolic values for those strings.
Although `lex' can be used by independently of `yacc', it is normally used to identify the different types of string which define the syntax of a file. For example, suppose `yacc' was parsing a C program. On the beginning of a line, it might expect to find either a variable name or a preprocessor symbol. A variable name is just a string consisting of characters from the set `0-9a-Z_', whereas a preprocessor command always starts with the character `#'. `yacc' passes control to `lex' which reads the file and matches the first object on the line. If it finds a variable, it returns to `yacc' a token which is a number or value corresponding to `variable'. Similarly, if it finds a preprocessor command, it returns a token for that. If it doesn't match either type it returns something else and `yacc' signals a syntax error.
Here is a `yacc' file which parses a file consisting of lines of the form a+b, where $a$ and $b$ are numbers -- any other syntax is incorrect. We could have used this later in the example program for the client-server example, See section Socket streams.
You can learn more about lex and yacc in "Lex and Yacc", J. Levine, T. Mason and D. Brown, O'Reilly and Assoc.
%{ /*******************************************************************/ /* */ /* PARSER for a + b protocol */ /* */ /* The section between the single %'s gets copied verbatim into */ /* the resulting C code yacc generates -- including this comment! */ /* */ /*******************************************************************/ #include <stdio.h> extern char *yytext; %} %token NUMBER PLUS %% specification: { yyerror("Warning: invalid statement");} | statement; statement: NUMBER PLUS NUMBER;
The lexer to go with this parser generates the tokens NUMBER and PLUS used by `yacc':
%{ /*******************************************************************/ /* */ /* LEXER for a + b protocol */ /* */ /* Returns token types NUMBER and PLUS to yacc, one at a time */ /* */ /*******************************************************************/ #include "y.tab.h" /* yacc produces this -- need this line! */ %} number [0-9]+ plus [+] %% number { return NUMBER; } plus { return PLUS; } . { return yytext[0]; } %% /* EOF */
The main program which uses `yacc' and `lex' looks like this:
extern FILE *yyin; main () { if ((yyin = fopen("My_Input_File","r")) == NULL) /* Open file */ { printf("Can't open file\n"); exit (1); } while (!feof(yyin)) { yyparse(); } fclose (yyin); }
Client-server communication is the basis of modern operating system technology. The UNIX socket mechanism makes stream-based communication virtually transparent.
Analogous to filestreams are sockets or TCP/IP network connections. A socket is a two-way (read/write) pseudo-file node. An open socket stream is like an open file-descriptor. Berkeley sockets are part of the standard C library.
There are two main kinds of socket: TCP/IP sockets and UNIX domain
sockets. UNIX sockets can be used to provide local interprocess
communication using a filestream communication protocol. TCP/IP sockets
open file descriptors across the network.
A TCP/IP socket is
a file stream associated with an IP address and a port number.
We write to a socket descriptor just as with a file descriptor, either
with write()
or using send()
.
When sending binary data over a network we have to be careful about machine level representations of 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.
----------------------------------- Big | 2 | 17 | 34 | 126 | ----------------------------------- ----------------------------------- Little | 126 | 34 | 17 | 2 | -----------------------------------
Obviously if we are 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. UNIX systems provide generic functions for converting between host-byteorder and network-byteorder for small and long integer data:
htonl, htons, ntohl, ntohs
Here we list two example programs which show how to make a client-server pair. The server enters a loop, and listens for connections from any clients (the generic address `INADDR_ANY' is a wildcard for any address on the current local network segment). The client program sends requests to the server as a protocol in the form of a string of the type `a + b'. Normally `a' and `b' are numbers, in which case the server returns their sum to the client. If the message has the special form `halt + *', where the star is arbitrary, then the server shuts down. Any other form of message results in an error, which the server signals to the client.
The basic structure of the client-server components in terms of system calls is this:
Client: socket() Create a socket connect() Contact a server socket (IP + port) while (?) { send() Send to server recv() Receive from server } Server: socket() Create a socket bind() Associates the socket with a fixed address listen() Create a listen queue while() { reply=accept() Accept a connection request recv() Receive from client send() Send to client }
/**********************************************************************/ /* */ /* The client part of a client-server pair. This simply takes two */ /* numbers and adds them together, returning the result to the client */ /* */ /* Compiled with: */ /* cc server.c */ /* */ /* User types: */ /* 3 + 5 */ /* a + b */ /* halt + server */ /**********************************************************************/ #include <stdio.h> #include <sys/types.h> #include <sys/socket.h> #include <netinet/in.h> #include <netdb.h> #define PORT 9000 /* Arbitrary non-reserved port */ #define HOST "nexus.iu.hio.no" #define bufsize 20 /**********************************************************************/ /* Main */ /**********************************************************************/ main (argc,argv) int argc; char *argv[]; { struct sockaddr_in cin; struct hostent *hp; char buffer[bufsize]; int sd; if (argc != 4) { printf("syntax: client a + b\n"); exit(1); } if ((hp = gethostbyname(HOST)) == NULL) { perror("gethostbyname: "); exit(1); } memset(&cin,0,sizeof(cin)); /* Another way to zero memory */ cin.sin_family = AF_INET; cin.sin_addr.s_addr = ((struct in_addr *)(hp->h_addr))->s_addr; cin.sin_port = htons(PORT); printf("Trying to connect to %s = %s\n",HOST,inet_ntoa(cin.sin_addr)); if ((sd = socket(AF_INET,SOCK_STREAM,0)) == -1) { perror("socket"); exit(1); } if (connect(sd,&cin,sizeof(cin)) == -1) { perror("connect"); exit(1); } sprintf(buffer,"%s + %s",argv[1],argv[3]); if (send(sd,buffer,strlen(buffer),0) == -1) { perror ("send"); exit(1); } if (recv(sd,buffer,bufsize,0) == -1) { perror("recv"); exit (1); } printf ("Server responded with %s\n",buffer); close (sd); unlink("./socket"); }
/**********************************************************************/ /* */ /* The server part of a client-server pair. This simply takes two */ /* numbers and adds them together, returning the result to the client */ /* */ /* Compiled with: */ /* cc server.c */ /* */ /**********************************************************************/ #include <stdio.h> #include <sys/types.h> #include <sys/socket.h> #include <netinet/in.h> #include <netdb.h> #define PORT 9000 #define bufsize 20 #define queuesize 5 #define true 1 #define false 0 /**********************************************************************/ /* Main */ /**********************************************************************/ main () { struct sockaddr_in cin; struct sockaddr_in sin; struct hostent *hp; char buffer[bufsize]; int sd, sd_client, addrlen; memset(&sin,0,sizeof(sin)); /* Another way to zero memory */ sin.sin_family = AF_INET; sin.sin_addr.s_addr = INADDR_ANY; /* Broadcast address */ sin.sin_port = htons(PORT); if ((sd = socket(AF_INET,SOCK_STREAM,0)) == -1) { perror("socket"); exit(1); } if (bind(sd,&sin,sizeof(sin)) == -1) /* Must have this on server */ { perror("bind"); exit(1); } if (listen(sd,queuesize) == -1) { perror("listen"); exit(1); } while (true) { if ((sd_client = accept(sd,&cin,&addrlen)) == -1) { perror("accept"); exit(1); } if (recv(sd_client,buffer,sizeof(buffer),0) == -1) { perror("recv"); exit(1); } if (!DoService(buffer)) { break; } if (send(sd_client,buffer,strlen(buffer),0) == -1) { perror("send"); exit(1); } close (sd_client); } close (sd); printf("Server closing down...\n"); } /**************************************************************/ DoService(buffer) char *buffer; /* This is the protocol section. Here we must */ /* check that the incoming data are sensible */ { int a=0,b=0; printf("Received: %s\n",buffer); sscanf(buffer,"%d + %d\n",&a,&b); if (a > 0 && b> 0) { sprintf(buffer,"%d + %d = %d",a,b,a+b); return true; } else { if (strncmp("halt",buffer,4) == 0) { sprintf(buffer,"Server closing down!"); return false; } else { sprintf(buffer,"Invalid protocol"); return true; } } }
In the example we use `streams' to implement a typical input/output behaviour for C. A stream interface is a so-called reliable protocol. There are other kinds of sockets too, called unrealiable, or UDP sockets. Features to notice on the server are that we must bind to a specific address. The client is always implicitly bound to an address since a socket connection always originates from the machine on which the client is running. On the server however we want to know which addresses we shall be receiving requests from. In the above example we use the generic wildcard address `INADDR_ANY' which means that any host can connect to the server. Had we been more specific, we could have limited communication to two machines only.
By calling `listen()' we set up a queue for incoming connections. Rather than forking a separate process to handle each request we set up a queue of a certain depth. If we exceed this depth then new clients rtying to connect will be refused connection.
The `accept' call is the mechanism which extracts a `reply handle' from the socket. Using the handle obtained from this call we can reply to the client without having to open a special socket explicitly.
An improved server side connection can be setup, reading the service name from `/etc/services' and setting reusable socket options to avoid busy signals, like this:
struct sockaddr_in cin, sin; struct servent *server; int sd, addrlen = sizeof(cin); int portnumber, yes=1; if ((server = getservbyname(service-name,"tcp")) == NULL) { CfLog(cferror,"Couldn't get cfengine service","getservbyname"); exit (1); } bzero(&cin,sizeof(cin)); /* Service returns network byte order */ sin.sin_port = (unsigned short)(server->s_port); sin.sin_addr.s_addr = INADDR_ANY; sin.sin_family = AF_INET; if ((sd = socket(AF_INET,SOCK_STREAM,0)) == -1) { CfLog(cferror,"Couldn't open socket","socket"); exit (1); } if (setsockopt (sd, SOL_SOCKET, SO_REUSEADDR, (char *) &yes, sizeof (int)) == -1) { CfLog(cferror,"Couldn't set socket options","sockopt"); exit (1); } if (bind(sd,(struct sockaddr *)&sin,sizeof(sin)) == -1) { } /* etc */
All the arguments must be collected into a struct, since only one argument pointer can be sent to the pthread functions.
#include <pthread.h> SpawnCfGetFile(args) struct cfd_thread_arg *args; { pthread_t tid; void *CfGetFile(); pthread_attr_init(&PTHREADDEFAULTS); pthread_attr_setdetachstate(&PTHREADDEFAULTS,PTHREAD_CREATE_DETACHED); if (pthread_create(&tid,&PTHREADDEFAULTS,CfGetFile,args) != 0) { CfLog(cferror,"pthread_create failed","create"); CfGetFile(args); } pthread_attr_destroy(&PTHREADDEFAULTS); } /***************************************************************/ void *CfGetFile(args) struct cfd_thread_arg *args; { pthread_mutex_t mutex; if (pthread_mutex_lock(&mutex) != 0) { CfLog(cferror,"pthread_mutex_lock failed","pthread_mutex_lock"); free(args->replyfile); /* from strdup in each thread */ DeleteConn(args->connect); free((char *)args); return NULL; } ACTIVE_THREADS++; /* Global variable */ if (pthread_mutex_unlock(&mutex) != 0) { CfLog(cferror,"pthread_mutex_unlock failed","unlock"); } /* send data */ if (pthread_mutex_lock(&mutex) != 0) { CfLog(cferror,"pthread_mutex_lock failed","pthread_mutex_lock"); return; } ACTIVE_THREADS--; if (pthread_mutex_unlock(&mutex) != 0) { CfLog(cferror,"pthread_mutex_unlock failed","unlock"); } #endif return NULL; }
The C library calls which query the databases are, amongst others,
getpwnam get password data by name getpwuid get password data by uid getgrnam get group data by name gethostent get entry in hosts database getnetgrent get entry in netgroups database getservbyname get servive by name getservbyport get service by port get protobyname get protocol by name
For a complete list and how to use these, see the UNIX manual.
The following example shows how to read the password file of the system. The functions used here can be used regardless of whether the network information service (NIS) is in use. The data are returned in a structure which is defined in `/usr/include/pwd.h'.
/******************************************************************/ /* */ /* Read the passwd file by name and sequentially */ /* */ /******************************************************************/ #include <unistd.h> #include <pwd.h> main () { uid_t uid; struct passwd *pw; uid = getuid(); pw = getpwuid(uid); printf ("Your login name is %s\n",pw->pw_name); printf ("Now here comes the whole file!\n\n"); setpwent(); while (getpwent()) { printf ("%s:%s:%s\n",pw->pw_name,pw->pw_gecos,pw->pw_dir); } endpwent(); }
The second network database service is that which converts host and domain names into IP numbers and vice versa. This is the domain name service, usually implemented by the BIND (Berkeley Internet Name Domain) software. The information here concerns version 4.9 of this software.
This is perhaps the most important function form hostname lookup. `gethostbyname()' gets its information either from files, NIS or DNS. Its behaviour is configured by the files mentioned above, See section DNS - The Domain Name Service. It is used to look up the IP address of a named host (including domain name if DNS is used). On the configurable systems described above, the full list of servers is queried until a reply is obtained. The order in which the different services are queried is important here since DNS returns a fully qualified name (host name plus domain name) whereas NIS and the `/etc/hosts' file database return only a hostname.
gethostbyname returns data in the form of a pointer to a static data structure. The syntax is
#include <netdb.h> struct hostent *hp; hp = gethostbyname("myhost.domain.country")
The resulting structure varies on different implementations of UNIX, but the `old BSD standard' is of the form:
struct hostent { char *h_name; /* official name of host */ char **h_aliases; /* alias list */ int h_addrtype; /* host address type */ int h_length; /* length of address */ char **h_addr_list; /* list of addresses from name server */ }; #define h_addr h_addr_list[0] /* address, for backward compatiblity */
The structure contains a list of addresses and or aliases from the nameserver. The interesting quantity is usually extracted by means of the macro `h_addr' whcih gives the first value in the address list, though officially one should examine the whole list now.
This value is a pointer which can be converted into a text form by the following hideous type transformation:
#include <sys/types.h> #include <sys/socket.h> #include <netinet/in.h> struct sockaddr_in sin; cin.sin_addr.s_addr = ((struct in_addr *)(hp->h_addr))->s_addr; printf("IP address = %s\n",inet_ntoa(cin.sin_addr));
See the client program in the first section of this chapter for an example of its use.
The support for NFS mounting in the standard C library is through two sources. NFS is based on the Sun's RPC system, so the basic calls are only instances of standard RPC protocols.
The C functions in the standard input/output library can be used to access NFS filesystems. Since NFS imitates the UNIX filesystem as closely as possible, NFS filesystems can be mounted in exactly the same way as ordinary filesystems. Unfortunately, the C functions which perform the mount operation in UNIX and depressingly non-standard. They differ on almost every implementation of UNIX.
The basic function which mounts a filesystem, in `mount' (see man (2) mount). The mount table is stored in a file /etc/mtab on BSD systems (again the name varies wildly from UNIX to UNIX, mnttab on HPUX for instance). The file /etc/rmtab on an NFS server contains a list of remote-mounted filesystems which are mounted by remote clients. C functions exist which can read the filesystem tables and place the resulting data in C struct types. Alas, these struct defintions are also quite different on different systems. See `/usr/include/sys/mount.h', so the user wishing to write system-independent code is confounded at the lowest level.
rls (options) hostname:/path/to/file
# C shell True - non-zero/non-empty value False - zero or null string # Bourne shell True - 0 returned by shell command False - non-zero returned by shell command ( Note that "test" converts from C shell style to Bourne shell) # Perl True - non-zero/non-empty value False - zero or null string /* C */ True - non zero integer False - zero integer
# C shell $< # Bourne shell line read # Perl <STDIN> /* C */ scanf
# C shell command > file command >& file command >> file command1 | command2 # Bourne shell command > file command > file 2>&1 command >> file command1 | command2 # Perl open (HANDLE,">file") open (HANDLE,">file 2>&1") open (HANDLE,">>file") open (HANDLE,"command1 |") open (HANDLE,"| command2") /* C */ fopen ("file","w"); printf(..) fopen ("file","w"); printf(..); fprintf(stderr,..) fopen ("file","a"); printf(..) popen ("command1","r") popen ("command2","w")
/* C */ Shell foreach end if then else endif while end switch case breaksw endsw repeat # Bourne shell while do done if then else fi until do done case in esac for in do done # Perl while if then else for unless else foreach until do while do until /* C */ while if then else do while switch case for
# C shell $argv[] $#argv # Bourne Shell $1, $2, $3... $* $# # Perl $ARGV[] $#ARGV /* C */ char argv[][] int argc
# C shell a = $b + $c # Bourne shell a = `expr $b + $c` # Perl $a = $b + $c; /* C */ a = b + c;
# C shell if ( $x == $y ) then endif # Bourne shell if [ $x -eq $y ]; then fi # Perl if ( $x == $y ) { } /* C */ if ( x == y ) { }
# C shell if ( $x == $y ) then endif # Bourne shell if [ $x = $y ]; then fi # Perl if ( $x eq $y ) then { } /* C */ if (strcmp(x,y) == 0) { }
# C shell, Bourne shell - cannot be done (pipes only) # Perl open (READ_HANDLE,"filename"); open (WRITE_HANDLE,"> filename"); open (APPEND_HANDLE,">> filename"); /* C */ FILE *fp; fp = fopen ("file","r"); fp = fopen ("file","w"); fp = fopen ("file","a");
# C shell foreach dir ( directory/* ) ... end # Bourne shell for dir in directory/* ; do ... done # Perl opendir (HANDLE,"directory") || die; while ($entry = readdir(HANDLE)) { } closedir(HANDLE); # C #include <dirent.h> DIR *dirh; struct dirent *dirp; if ((dirh = opendir(name)) == NULL) { perror("opendir") exit(1); } for (dirp = readdir(dirh); dirp != NULL; dirp = readdir(dirh)) { ... /* dirp->d_name points to child */ } closedir(dirh);
# C shell if ( -f file ) # plain file if ( -d file ) # directory # Bourne shell if [ -f file ] # plain file if [ -d file ] # directory # Perl if ( -f file ) # plain file if ( -d file ) # directory if ( -l file ) # symbolic link /* C */ #include <sys/stat.h> struct stat statvar; stat("file", &statvar); if (S_ISREG(statvar.mode)) /* plain file */ if (S_ISDIR(statvar.mode)) /* directory */ lstat("file", &statvar); if (S_ISLNK(statvar.mode)) /* symbolic link */
argc
in C
argv
in C
crypt()
fork()
getenv()
ln
ln -s
rand()
cc
cc
accept()
bind()
chgrp
command
chmod
command
chown
command
closedir
command
connect()
egrep
command
env
command
envp
in C
extern
variables
for
loop in Bash
for
loop in perl
foreach
foreach
example
foreach
loop in perl
getenv() function
getgrnam()
gethostbyname()
, gethostbyname()
gethostent()
getnetgrent()
getpwnam()
getpwuid()
getservbyname()
getservbyport()
if..then..else
in csh
if..then..else..fi
in Bash
INADDR_ANY
ioctl()
listen()
ln -s
ls -l
lstat()
noclobber
overwrite protection
noclobber
variable
opendir
command
PATH
path
popen()
readdir
command
readlink()
recv()
repeat
send()
set
command
setenv
command
shift
operator on strings
socket()
stat()
switch..case
in csh
umask
variable, umask
variable
unset
command
while
while
loop in Bash
write
example
WTERMSIG(status)
This document was generated on 23 August 2001 using the texi2html translator version 1.51.