writing more secure perl cgi scripts
1.0.0
How to write more secure Perl CGI scripts. A guide with some tips.
Update March 2020: Migrate this tutorial from 2008 to Github for archival purposes. (Nowadays there are better programming languages and web frameworks).
This guide contains some example CGI scripts that demonstrates how to improve security in CGI scripts. The example count7.cgi has the best security.
If you are writing your CGI scripts in perl, you should
use the -T flag. It is also a good idea to use the -w flag. The script will hence start with
#!/usr/bin/perl -Tw
Avoid using temporary files as much as possible. If you want to call an external program, try to use the open2() call. This works under the assumption that the external program can read the input data from standard input and write the result to standard output. For instance, like this
use IPC::Open2;
my $childpid = open2(*READ, *WRITE, $config::blastall_path , "-m", "7", "-d" , $db, "-p", "blastp", "-b", $max_hits, "-v", $max_hits, "-M", "BLOSUM62" )
or die "Could not open pipe: $!";
print WRITE $blastinput;
close(WRITE);
my $blastoutput;
while(<READ>) {
$blastoutput .= $_;
}
close READ;
waitpid($childpid, 0);
never use something like this
system("myprogram -f $cgiparam");
instead use the comma separated syntax
system("myprogram","-f","$cgiparam");
count.py is a Python script that is used by all of the example cgi scripts.
#!/usr/bin/python
import sys
if len(sys.argv) == 2:
sys.exit()
elif len(sys.argv) == 3:
f=open(sys.argv[2], "r")
else:
sys.exit()
#f.seek(0)
lines=0
bytes=0
for line in f:
lines += 1
bytes += len( line )
if (sys.argv[1] == "bytes"):
print str(bytes)
if (sys.argv[1] == "lines"):
print str(lines)
Configuring apache httpd to allow Server Side Includes ( SSI ) is dangerous. The option is called Includes.
A hacker could upload a file with SSI exec statements. Those commands would then be executed as soon as a web browser tries to access the uploaded file. Often SSI is only activated for filenames ending with ".shtml". In the this CGI script the web surfer chooses the file name and can choose a file name ending with ".shtml".
#!/usr/bin/perl
use CGI qw(:standard);
use CGI::Carp qw(fatalsToBrowser);
my $cgi = new CGI;
if ($cgi->cgi_error()) { die() };
print
$cgi->header(),
$cgi->start_html( '' ),
$cgi->h1('Count'),
$cgi->start_form(-method=>'GET'),
"type:",
$cgi->popup_menu(-name=>'type',-values=>['lines','bytes']),
$cgi->p,
"filename",
$cgi->textfield(-name=>'fname'),
$cgi->p,
$cgi->textarea(-name=>'content', -rows=>10, -columns=>50),
$cgi->p,
$cgi->submit(),
$cgi->end_form,
$cgi->hr;
if ($cgi->param) {
my $fname = $cgi->param('fname');
my $content = $cgi->param('content');
my $type = $cgi->param('type');
my $filename="tmp/" . $fname;
my $outfh;
open($outfh, ">", $filename)
or die "Couldn't open $tmpfile for writing: $!";
print $outfh $content;
close $outfh;
my $result=`/home/esjolund/public_html/cgi-bin/count.py $type $filename`;
print "The file <b>$fname</b> has $result $type";
print $cgi->end_html;
}
Examplifies use File::Temp qw(tempfile);. If you can't avoid using temporary files, please use this perl module to create them. Although not possible here, you should preferrably use UNLINK => 1 as argument to the tempfile() function.
#!/usr/bin/perl
use CGI qw(:standard);
use CGI::Carp qw(fatalsToBrowser);
use File::Temp qw(tempfile);
my $cgi = new CGI;
if ($cgi->cgi_error()) { die() };
print
$cgi->header(),
$cgi->start_html( '' ),
$cgi->h1('Count'),
$cgi->start_form(-method=>'GET'),
"type:",
$cgi->popup_menu(-name=>'type',-values=>['lines','bytes']),
$cgi->p,
$cgi->textarea(-name=>'content', -rows=>10, -columns=>50),
$cgi->p,
$cgi->submit(),
$cgi->end_form,
$cgi->hr;
if ($cgi->param) {
my $content = $cgi->param('content');
my $type = $cgi->param('type');
File::Temp->safe_level( File::Temp::HIGH );
my ( $outfh, $filename ) =
tempfile( "tmp/myfiles.XXXXXX", UNLINK => 0 );
if ( !defined $outfh ) {
die "error: no temporary filen";
}
print $outfh $content;
close $outfh;
my $result=`/home/esjolund/public_html/cgi-bin/count.py $type $filename`;
print "Content has $result $type";
print $cgi->end_html;
}
It is often possible to avoid temporary files by using pipes. In Perl this is done with open2()
#!/usr/bin/perl
use CGI qw(:standard);
use CGI::Carp qw(fatalsToBrowser);
use IPC::Open2;
my $cgi = new CGI;
if ($cgi->cgi_error()) { die() };
print
$cgi->header(),
$cgi->start_html( '' ),
$cgi->h1('Count'),
$cgi->start_form(-method=>'GET'),
"type:",
$cgi->popup_menu(-name=>'type',-values=>['lines','bytes']),
$cgi->p,
$cgi->textarea(-name=>'content', -rows=>10, -columns=>50),
$cgi->p,
$cgi->submit(),
$cgi->end_form,
$cgi->hr;
if ($cgi->param) {
my $content = $cgi->param('content');
my $type = $cgi->param('type');
my($child_out, $child_in);
$pid = open2($child_out, $child_in, "/home/esjolund/public_html/cgi-bin/count.py", $type,"/dev/stdin");
print $child_in $content;
close($child_in);
my $result=<$child_out>;
waitpid($pid,0);
print "Content has $result $type";
print $cgi->end_html;
}
The previous example count3.cgi is insecure as it passes the $type variable as directly to the command line ( to a shell ). We should instead use the comma separated syntax of open2() as shown here:
#!/usr/bin/perl
use CGI qw(:standard);
use CGI::Carp qw(fatalsToBrowser);
use IPC::Open2;
my $cgi = new CGI;
if ($cgi->cgi_error()) { die() };
print
$cgi->header(),
$cgi->start_html( '' ),
$cgi->h1('Count'),
$cgi->start_form(-method=>'GET'),
"type:",
$cgi->popup_menu(-name=>'type',-values=>['lines','bytes']),
$cgi->p,
$cgi->textarea(-name=>'content', -rows=>10, -columns=>50),
$cgi->p,
$cgi->submit(),
$cgi->end_form,
$cgi->hr;
if ($cgi->param) {
my $content = $cgi->param('content');
my $type = $cgi->param('type');
my($child_out, $child_in);
$pid = open2($child_out, $child_in, "/home/esjolund/public_html/cgi-bin/count.py",$type);
print $child_in $content;
close($child_in);
my $result=<$child_out>;
waitpid($pid,0);
print "Content has $result $type";
print $cgi->end_html;
}
Examplifies taint mode, the -T flag
#!/usr/bin/perl -T
use CGI qw(:standard);
use CGI::Carp qw(fatalsToBrowser);
use IPC::Open2;
use Scalar::Util qw(tainted);
my $cgi = new CGI;
if ($cgi->cgi_error()) { die() };
print
$cgi->header(),
$cgi->start_html( '' ),
$cgi->h1('Count'),
$cgi->start_form(-method=>'GET'),
"type:",
$cgi->popup_menu(-name=>'type',-values=>['lines','bytes']),
$cgi->p,
$cgi->textarea(-name=>'content', -rows=>10, -columns=>50),
$cgi->p,
$cgi->submit(),
$cgi->end_form,
$cgi->hr;
if ($cgi->param) {
if (tainted($cgi->param('content')))
{ print 'variable content tainted!'; }
else
{ print 'variable content not tainted!'; }
print `echo $cgi->param('content')`;
my $content = $cgi->param('content');
my $type = $cgi->param('type');
my($child_out, $child_in);
$pid = open2($child_out, $child_in, "/home/esjolund/public_html/cgi-bin/count.py",$type);
print $child_in $content;
close($child_in);
my $result=<$child_out>;
waitpid($pid,0);
print "Content has $result $type";
print $cgi->end_html;
}
If we try to run the previous example count5.cgi, perl will complain that the PATH variable is insecure. If we use the -T flag, we need to set the PATH environment variable.
#!/usr/bin/perl -T
use CGI qw(:standard);
use CGI::Carp qw(fatalsToBrowser);
use IPC::Open2;
use Scalar::Util qw(tainted);
my $cgi = new CGI;
if ($cgi->cgi_error()) { die() };
$ENV{'PATH'} = '/bin:/usr/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};
print
$cgi->header(),
$cgi->start_html( '' ),
$cgi->h1('Count'),
$cgi->start_form(-method=>'GET'),
"type:",
$cgi->popup_menu(-name=>'type',-values=>['lines','bytes']),
$cgi->p,
$cgi->textarea(-name=>'content', -rows=>10, -columns=>50),
$cgi->p,
$cgi->submit(),
$cgi->end_form,
$cgi->hr;
if ($cgi->param) {
my $content = $cgi->param('content');
my $type = $cgi->param('type');
if ( tainted($type) )
{ print 'tainted!'; }
else
{ print 'not tainted!'; }
print $cgi->hr;
print `echo $type`;
my($child_out, $child_in);
$pid = open2($child_out, $child_in, "/home/esjolund/public_html/cgi-bin/count.py",$type);
print $child_in $content;
close($child_in);
my $result=<$child_out>;
waitpid($pid,0);
print "Content has $result $type";
print $cgi->end_html;
}
If we try to run the previous example count6.cgi perl will complain that we are using untainted variables in an insecure way. To use CGI input variables we need to untaint them by processing them with a regular expression
#!/usr/bin/perl -T
use CGI qw(:standard);
use CGI::Carp qw(fatalsToBrowser);
use IPC::Open2;
use Scalar::Util qw(tainted);
my $cgi = new CGI;
if ($cgi->cgi_error()) { die() };
$ENV{'PATH'} = '/bin:/usr/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};
print
$cgi->header(),
$cgi->start_html( '' ),
$cgi->h1('Count'),
$cgi->start_form(-method=>'GET'),
"type:",
$cgi->popup_menu(-name=>'type',-values=>['lines','bytes']),
$cgi->p,
$cgi->textarea(-name=>'content', -rows=>10, -columns=>50),
$cgi->p,
$cgi->submit(),
$cgi->end_form,
$cgi->hr;
if ($cgi->param) {
my $type;
if ( $cgi->param('type')=~/^(bytes|lines)$/ ) {
$type = $1;
} else {
print "variable type needs to be bytes or lines";
exit(0);
}
my $content;
if ( $cgi->param('content')=~/^(.*)$/s ) {
$content = $1;
} else {
exit(0);
}
if ( tainted($type) )
{ print 'tainted!'; }
else
{ print 'not tainted!'; }
print $cgi->hr;
print `echo $type`;
my($child_out, $child_in);
$pid = open2($child_out, $child_in, "/home/esjolund/public_html/cgi-bin/count.py",$type);
print $child_in $content;
close($child_in);
my $result=<$child_out>;
waitpid($pid,0);
print "Content has $result $type";
print $cgi->end_html;
}
