As you know from basic shell usage, input and output of a command may be redirected before it is executed, using a special notation - the redirection operators - interpreted by the shell. Redirection may also be used to open and close files for the current shell execution environment.

Redirection can also occur in a script, so that it can receive input from a file, for instance, or send output to a file. Later, the user can review the output file, or it may be used by another script as input.

File input and output are accomplished by integer handles that track all open files for a given process. These numeric values are known as file descriptors. The best known file descriptors are stdin, stdout and stderr, with file descriptor numbers 0, 1 and 2, respectively. These numbers and respective devices are reserved. Bash can take TCP or UDP ports on networked hosts as file descriptors as well.

The output below shows how the reserved file descriptors point to actual devices:

michel ~> ls -l /dev/std*
lrwxrwxrwx  1 root    root     17 Oct  2 07:46 /dev/stderr -> ../proc/self/fd/2
lrwxrwxrwx  1 root    root     17 Oct  2 07:46 /dev/stdin -> ../proc/self/fd/0
lrwxrwxrwx  1 root    root     17 Oct  2 07:46 /dev/stdout -> ../proc/self/fd/1

michel ~> ls -l /proc/self/fd/[0-2]
lrwx------  1 michel  michel   64 Jan 23 12:11 /proc/self/fd/0 -> /dev/pts/6
lrwx------  1 michel  michel   64 Jan 23 12:11 /proc/self/fd/1 -> /dev/pts/6
lrwx------  1 michel  michel   64 Jan 23 12:11 /proc/self/fd/2 -> /dev/pts/6

Note that each process has its own view of the files under /proc/self, as it is actually a symbolic link to /proc/<process_ID>.

You might want to check info MAKEDEV and info proc for more information about /proc subdirectories and the way your system handles standard file descriptors for each running process.

When excuting a given command, the following steps are excuted, in order:

When you run a script from the command line, nothing much changes because the child shell process will use the same file descriptors as the parent. When no such parent is available, for instance when you run a script using the cron facility, the standard file descriptors are pipes or other (temporary) files, unless some form of redirection is used. This is demonstrated in the example below, which shows output from a simple at script:

michel ~> date
Fri Jan 24 11:05:50 CET 2003

michel ~> at 1107
warning: commands will be executed using (in order) 
a) $SHELL b) login shell c)/bin/sh
at> ls -l /proc/self/fd/ > /var/tmp/fdtest.at
at> <EOT>
job 10 at 2003-01-24 11:07

michel ~> cat /var/tmp/fdtest.at
total 0
lr-x------    1 michel michel  64 Jan 24 11:07 0 -> /var/spool/at/!0000c010959eb (deleted)
l-wx------    1 michel michel  64 Jan 24 11:07 1 -> /var/tmp/fdtest.at
l-wx------    1 michel michel  64 Jan 24 11:07 2 -> /var/spool/at/spool/a0000c010959eb
lr-x------    1 michel michel  64 Jan 24 11:07 3 -> /proc/21949/fd

And one with cron:

michel ~> crontab -l
# DO NOT EDIT THIS FILE - edit the master and reinstall.
# (/tmp/crontab.21968 installed on Fri Jan 24 11:30:41 2003)
# (Cron version -- $Id$)
32 11 * * * ls -l /proc/self/fd/ > /var/tmp/fdtest.cron

michel ~> cat /var/tmp/fdtest.cron
total 0
lr-x------    1 michel michel  64 Jan 24 11:32 0 -> pipe:[124440]
l-wx------    1 michel michel  64 Jan 24 11:32 1 -> /var/tmp/fdtest.cron
l-wx------    1 michel michel  64 Jan 24 11:32 2 -> pipe:[124441]
lr-x------    1 michel michel  64 Jan 24 11:32 3 -> /proc/21974/fd

From the previous examples, it is clear that you can provide input and output files for a script (see Section 2.4, “File input and output” for more), but some tend to forget about redirecting errors - output which might be depended upon later on. Also, if you are lucky, errors will be mailed to you and eventual causes of failure might get revealed. If you are not as lucky, errors will cause your script to fail and won't be caught or sent anywhere, so that you can't start to do any worthwhile debugging.

When redirecting errors, note that the order of precedence is significant. For example, this command, issued in /var/spool

ls -l * 2> /var/tmp/unaccessible-in-spool

will redirect standard output of the ls command to the file unaccessible-in-spool in /var/tmp. The command

ls -l * > /var/tmp/spoollist 2>&1

will direct both standard input and standard error to the file spoollist. The command

ls -l * 2 >& 1 > /var/tmp/spoollist

directs only the standard output to the destination file, because the standard error is copied to standard output before the standard output is redirected.

For convenience, errors are often redirected to /dev/null, if it is sure they will not be needed. Hundreds of examples can be found in the startup scripts for your system.

Bash allows for both standard output and standard error to be redirected to the file whose name is the result of the expansion of FILE with this construct:

This is the equivalent of > FILE 2>&1, the construct used in the previous set of examples. It is also often combined with redirection to /dev/null, for instance when you just want a command to execute, no matter what output or errors it gives.

The /dev/fd directory contains entries named 0, 1, 2, and so on. Opening the file /dev/fd/N is equivalent to duplicating file descriptor N. If your system provides /dev/stdin, /dev/stdout and /dev/stderr, you will see that these are equivalent to /dev/fd/0, /dev/fd/1 and /dev/fd/2, respectively.

The main use of the /dev/fd files is from the shell. This mechanism allows for programs that use pathname arguments to handle standard input and standard output in the same way as other pathnames. If /dev/fd is not available on a system, you'll have to find a way to bypass the problem. This can be done for instance using a hyphen (-) to indicate that a program should read from a pipe. An example:

michel ~> filter body.txt.gz | cat header.txt - footer.txt
This text is printed at the beginning of each print job and thanks the sysadmin
for setting us up such a great printing infrastructure.

Text to be filtered.

This text is printed at the end of each print job.

The cat command first reads the file header.txt, next its standard input which is the output of the filter command, and last the footer.txt file. The special meaning of the hyphen as a command-line argument to refer to the standard input or standard output is a misconception that has crept into many programs. There might also be problems when specifying hyphen as the first argument, since it might be interpreted as an option to the preceding command. Using /dev/fd allows for uniformity and prevents confusion:

michel ~> filter body.txt | cat header.txt /dev/fd/0 footer.txt | lp

In this clean example, all output is additionally piped through lp to send it to the default printer.

Since child processes inherit open file descriptors, it is good practice to close a file descriptor when it is no longer needed. This is done using the

syntax. In the above example, file descriptor 7, which has been assigned to standard input, is closed each time the user needs to have access to the actual standard input device, usually the keyboard.

The following is a simple example redirecting only standard error to a pipe:

michel ~> cat listdirs.sh
#!/bin/bash

# This script prints standard output unchanged, while standard error is 
# redirected for processing by awk.

INPUTDIR="$1"

# fd 6 targets fd 1 target (console out) in current shell
exec 6>&1

# fd 1 targets pipe, fd 2 targets fd 1 target (pipe),
# fd 1 targets fd 6 target (console out), fd 6 closed, execute ls
ls "$INPUTDIR"/* 2>&1 >&6 6>&- \
				# Closes fd 6 for awk, but not for ls.

| awk 'BEGIN { FS=":" } { print "YOU HAVE NO ACCESS TO" $2 }' 6>&-

# fd 6 closed for current shell
exec 6>&-

Frequently, your script might call on another program or script that requires input. The here document provides a way of instructing the shell to read input from the current source until a line containing only the search string is found (no trailing blanks). All of the lines read up to that point are then used as the standard input for a command.

The result is that you don't need to call on separate files; you can use shell-special characters, and it looks nicer than a bunch of echo's:

michel ~> cat startsurf.sh
#!/bin/bash

# This script provides an easy way for users to choose between browsers.

echo "These are the web browsers on this system:"
 
# Start here document
cat << BROWSERS
mozilla
links
lynx
konqueror
opera
netscape
BROWSERS
# End here document

echo -n "Which is your favorite? "
read browser

echo "Starting $browser, please wait..."
$browser &

michel ~> startsurf.sh
These are the web browsers on this system:
mozilla
links
lynx
konqueror
opera
netscape
Which is your favorite? opera
Starting opera, please wait...

Although we talk about a here document, it is supposed to be a construct within the same script. This is an example that installs a package automatically, eventhough you should normally confirm:

#!/bin/bash
 
# This script installs packages automatically, using yum.
 
if [ $# -lt 1 ]; then
        echo "Usage: $0 package."
        exit 1
fi
 
yum install $1 << CONFIRM
y
CONFIRM

And this is how the script runs. When prompted with the “Is this ok [y/N]” string, the script answers “y” automatically:

[root@picon bin]# ./install.sh tuxracer
Gathering header information file(s) from server(s)
Server: Fedora Linux 2 - i386 - core
Server: Fedora Linux 2 - i386 - freshrpms
Server: JPackage 1.5 for Fedora Core 2
Server: JPackage 1.5, generic
Server: Fedora Linux 2 - i386 - updates
Finding updated packages
Downloading needed headers
Resolving dependencies
Dependencies resolved
I will do the following:
[install: tuxracer 0.61-26.i386]
Is this ok [y/N]: EnterDownloading Packages
Running test transaction:
Test transaction complete, Success!
tuxracer 100 % done 1/1
Installed:  tuxracer 0.61-26.i386
Transaction(s) Complete