It's hard to write a bash script of much import without using a pipe or two. Named pipes, on the other hand, are much rarer.

Like un-named/anonymous pipes, named pipes provide a form of IPC (Inter-Process Communication). With anonymous pipes, there's one reader and one writer, but that's not required with named pipes—any number of readers and writers may use the pipe.

Named pipes are visible in the filesystem and can be read and written just as other files are:

$ ls -la /tmp/testpipe
prw-r--r-- 1 mitch users 0 2009-03-25 12:06 /tmp/testpipe|

Why might you want to use a named pipe in a shell script? One situation might be when you've got a backup script that runs via cron, and after it's finished, you want to shut down your system. If you do the shutdown from the backup script, cron never sees the backup script finish, so it never sends out the e-mail containing the output from the backup job. You could do the shutdown via another cron job after the backup is "supposed" to finish, but then you run the risk of shutting down too early every now and then, or you have to make the delay much larger than it needs to be most of the time.

Using a named pipe, you can start the backup and the shutdown cron jobs at the same time and have the shutdown just wait till the backup writes to the named pipe. When the shutdown job reads something from the pipe, it then pauses for a few minutes so the cron e-mail can go out, and then it shuts down the system.

Of course, the previous example probably could be done fairly reliably by simply creating a regular file to signal when the backup has completed. A more complex example might be if you have a backup that wakes up every hour or so and reads a named pipe to see if it should run. You then could write something to the pipe each time you've made a lot of changes to the files you want to back up. You might even write the names of the files that you want backed up to the pipe so the backup doesn't have to check everything.

Named pipes are created via mkfifo or mknod:

$ mkfifo /tmp/testpipe
$ mknod /tmp/testpipe p

The following shell script reads from a pipe. It first creates the pipe if it doesn't exist, then it reads in a loop till it sees "quit":

#!/bin/bash

pipe=/tmp/testpipe

trap "rm -f $pipe" EXIT

if [[ ! -p $pipe ]]; then
    mkfifo $pipe
fi

while true
do
    if read line <$pipe; then
        if [[ "$line" == 'quit' ]]; then
            break
        fi
        echo $line
    fi
done

echo "Reader exiting"

The following shell script writes to the pipe created by the read script. First, it checks to make sure the pipe exists, then it writes to the pipe. If an argument is given to the script, it writes it to the pipe; otherwise, it writes "Hello from PID".

#!/bin/bash

pipe=/tmp/testpipe

if [[ ! -p $pipe ]]; then
    echo "Reader not running"
    exit 1
fi


if [[ "$1" ]]; then
    echo "$1" >$pipe
else
    echo "Hello from $$" >$pipe
fi

Running the scripts produces:

$ sh rpipe.sh &
[3] 23842
$ sh wpipe.sh
Hello from 23846
$ sh wpipe.sh
Hello from 23847
$ sh wpipe.sh
Hello from 23848
$ sh wpipe.sh quit
Reader exiting

Note: initially I had the read command in the read script directly in the while loop of the read script, but the read command would usually return a non-zero status after two or three reads causing the loop to terminate.

while read line <$pipe
do
    if [[ "$line" == 'quit' ]]; then
        break
    fi
    echo $line
done