empty − run processes under pseudo-terminal sessions


empty −f [−i fifo1 −o fifo2] [-p] [−L file.log] command [command args]
−w [−Sv] [−t n] [−i fifo2 −o fifo1] key1 [answer1] ... [keyX answerX]
−s [−Sc] −o fifo1 [request]
−r [−b size] [−t n] [−i fifo2]
−k [pid] [signal]


empty is an utility that provides a simple interface to execute and/or interact with processes under pseudo-terminal sessions. This tool is definitely useful in programming of shell scripts which are used to communicate with interactive programs like telnet or ftp. In some cases empty can be a substitution for TCL/expect or other similar programming tools.

There are several common forms of command lines for empty. But the first execution of empty is usually a start in the daemon mode to fork a new command (process) under PTY-session. This can be done with −f key. An interface for the input and output data channels of the forked process is performed by two fifo files which names may be specified with −i and −o keys. These files are automatically created/deleted any time you start/exit empty daemon, so you must not create them manually. If you did not specify these fifo files in the command line, empty names them by itself basing on its PID and PID of forked PTY process.

At this point any application can easily communicate with forked process by writing data to the input fifo and reading answers from the output fifo, see EXAMPLES section for the details. To simplify this operations, empty offers an interface to just send any data (use −s key), or even to watch the output fifo for multiple keyphrases and reply to the input fifo with one of the responses (see −w key).

Note! Input fifo for empty -f ... is usually an output fifo for empty -w and empty -s forms. And output fifo of empty -f ... is an input fifo for empty -w ...

If something goes wrong the forked process may be killed by the standard kill command, or using −k key of empty. See −p option to save PID of empty daemon process.

The following options are available:


fork, spawn, start or execute a new process specified by the command and its arguments. If you omit fifo files, empty with its job control algorithm will create them under /tmp directory using this templates: and empty.PPID.PID.out, here PPID is usually your shell system process ID and PID is system process ID of empty-daemon.


send data (request) to the forked process. If fifo file was not specified with −o key, empty will try to find an automatically created fifo file it in /tmp directory. Instead of command line you can send your request or data directly to standard input (stdin) of empty


watch for one or more keyphrases and if specified send the appropriated response to the input fifo. If response is not set, empty waits for the proper keyphrase then exits. With −w key empty returns the number of matched keyphrase-response pair, or 255 if fails to find this match (see -t key for details of possible exit on timeout).


read from output FIFO one line (default) or one block of data (if −b size was specified). If −t n key was placed, exit on timeout.


list automatically created jobs by your shell. NB! Your custom jobs, which fifo files you specified with −i and −o keys, are not displayed. So if you did not specify fifo files with -i and -o keys all operations are done under the job marked current


send signal to the process with pid. If you did not specify pid, empty tries to find it within the list of automatically created jobs. If signal is omitted the default SIGTERM is used.


print short help message and exit

−i fifo1

a fifo file, which is used as input for a forked process.

−o fifo2

a fifo file, which is used as output for a forked process.

−L file.log

This option allows to log the whole empty session to a file. Marks >>> and <<< show the directions of data flow.


This option allows to choose the mode of the log file created with the -L option. The default mode is 0600.


Save PID of empty daemon process to a file

−t n

If input FIFO is empty, wait for n seconds (default is 10) to receive the keyphrase then exit on timeout with 255 code.


force empty to use stdin for data or requests.


Strip the last character from the input. Works with -s and -w keys


kvazi verbose mode. Show all contents of received buffer.


Start a new PTY-session with telnet to localhost:
empty -f -i in.fifo -o out.fifo -p -L empty.log telnet localhost
Interact with telnet:
empty −w −i out.fifo −o in.fifo ogin ’my_user\n’

empty −w −i out.fifo −o in.fifo assword ’my_password\n’

Send commands to telnet with empty:

empty −s −o in.fifo who

empty −s −o in.fifo "ls −la /\n"

The same using STDIN:

echo who | empty −s −o in.fifo

echo "ls −la /" | empty −s −o in.fifo

Just cat output from telnet:

cat out.fifo

Read one line from out.fifo:

empty −r -i out.fifo

Send commands to telnet with ordinary echo:

echo "who am i" > in.fifo

echo "uname -a" > in.fifo

Kill a process with PID 1234:

empty -k 1234


kill 1234

Telnet session with automatically created jobs:

empty -f telnet localhost

Interact with telnet using job control:

empty −w ogin ’my_user\n’

empty −w assword ’my_password\n’

List automatically created jobs:

empty −l


479 706 in /tmp/

479 706 out /tmp/empty.479.706.out

479 711 in /tmp/

479 711 out /tmp/empty.479.711.out

479 711 current


It is considered insecure to send a password in the command line like this:

empty −w assword ’my_password\n’

or like this:

empty −s ’my_password\n’

The reason is that the command line arguments are visible to the system while empty is running. Any local user can see them with ps(1), sometimes they are visible even remotely with finger(1). Also your server may have some monitoring tools which may store the output from ps(1) in their logs. There are also other, more complicated ways to compromise this information. Generally, you should take command line arguments as (possibly) visible to every one unless you really know what you’re doing.

empty with ’-s’ flag runs quickly in most cases, but still it can hang for a number of reasons (like fifo overloading), and even if it runs quick you still cannot be sure that no one will see its command line arguments even in this short time. empty with ’-w’ flag is even worse because it must wait for the keyphrase.

A better way to send the password to the supervised program is to read it from file:

empty −s [common options] <./password-file

or from a pipe:

get-password-of-user "$user" |empty -s [common options]

You should still make sure that you do not send any password via command line while creating this file, and certainly you should set some safe permissions to this file AND its directory (with the parent directories) before reading the password from the file OR writing the password to it.

Another possible way is to use your shell’s builtin (but see below):

echo "$password" |empty −s [common options]

Many shells like bash(1), csh(1) and FreeBSD’s sh(1) do not call external echo(1) command but use their own builtin echo command. Since no external command is started (the shell itself does all that echo(1) must do), nothing is shown in the process list. It is beyond this manual page to discuss the way to make sure that your shell uses the builtin command.


If any error occurs empty usually exits with code 255. Otherwise zero or some positive value (see −w key) is returned.


expect(1), chat(8)


empty was made by Mikhail E. Zakharov. This software was based on the basic idea of pty version 4.0 Copyright (c) 1992, Daniel J. Bernstein but no code was ported from pty4. SECURITY section of this manual page was contributed by Sergey Redin.