=head1 NAME

Net::LDAP::Security - Security issues with LDAP connections

=head1 SYNOPSIS

  none

=head1 DESCRIPTION

This document discusses various security issues relating to using LDAP
and connecting to LDAP servers, notably how to manage these potential
vulnerabilities:

=over 4

=item *

do you know that you are connected to the right server

=item *

can someone sniff your passwords/userids from the directory connection

=item *

can someone sniff other confidential information from the directory
connection

=back

B<Net::LDAP> provides ways to address these vulnerabilities: through the
use of LDAPS, or LDAPv3 and TLS, and/or the use of SASL. Each of these
will be explained below.

=head2 How does an LDAP connection work

A normal LDAPv2 or LDAPv3 connection works by the client connecting
directly to port 389 (by default), and then issuing various LDAP
requests like search, add, etc.

There is no way to guarantee that an LDAP client is connected to the
right LDAP server. Hackers could have poisoned your DNS, so
'ldap.example.com' could be made to point to 'ldap.hacker.com'. Or
they could have installed their own server on the correct machine.

It is in the nature of the LDAP protocol that all information goes
between the client and the server in 'plain text'. This is a term used
by cryptographers to describe unencrypted and recoverable data, so
even though LDAP can transfer binary values like JPEG photographs,
audio clips and X.509 certificates, everything is still considered
'plain text'.

If these vulnerabilities are an issue to, then you should consider the
other possibilities described below, namely LDAPS, LDAPv3 and TLS, and
SASL.

=head2 How does an LDAPS connection work

LDAPS is an unofficial protocol. It is to LDAP what HTTPS is to HTTP,
namely the exact same protocol (but in this case LDAPv2 or LDAPv3)
running over a I<secured> SSL ("Secure Socket Layer") connection to
port 636 (by default).

Not all servers will be configured to listen for LDAPS connections,
but if they do, it will commonly be on a different port from the normal
plain text LDAP port.

Using LDAPS can I<potentially> solve the vulnerabilities described
above, but you should be aware that simply "using" SSL is not a magic
bullet that automatically makes your system "secure".

First of all, LDAPS can solve the problem of verifying that you are
connected to the correct server. When the client and server connect,
they perform a special SSL 'handshake', part of which involves the
server and client exchanging cryptographic keys, which are described
using X.509 certificates. If the client wishes to confirm that it is
connected to the correct server, all it needs to do is verify the
server's certificate which is sent in the handshake. This is done in
two ways:

=over 4

=item 1

check that the certificate is signed (trusted) by someone that you
trust, and that the certificate hasn't been revoked. For instance, the
server's certificate may have been signed by Verisign
(www.verisign.com), and you decide that you want to trust Verisign to
sign legitimate certificates.

=item 2

check that the least-significant cn RDN in the server's certificate's
DN is the fully-qualified hostname of the hostname that you connected
to when creating the LDAPS object. For example if the server is
E<lt>cn=ldap.example.com,ou=My department,o=My companyE<gt>, then the
RDN to check is cn=ldap.example.com.

=back

You can do this by using the cafile and capath options when creating a
B<Net::LDAPS> object, I<and> by setting the verify option to 'require'.

To prevent hackers 'sniffing' passwords and other information on your
connection, you also have to make sure the encryption algorithm used
by the SSL connection is good enough. This is also something that gets
decided by the SSL handshake - if the client and server cannot agree
on an acceptable algorithm the connection is not made.

B<Net::LDAPS> will by default use all the algorithms built into your copy
of OpenSSL, except for ones considered to use "low" strength
encryption, and those using export strength encryption. You can
override this when you create the B<Net::LDAPS> object using the
'ciphers' option.

Once you've made the secure connection, you should also check that the
encryption algorithm that is actually being used is one that you find
acceptable. Broken servers have been observed in the field which 'fail
over' and give you an unencrypted connection, so you ought to check
for that.

=head2 How does LDAP and TLS work

SSL is a good solution to many network security problems, but it is
not a standard. The IETF corrected some defects in the SSL mechanism
and published a standard called RFC 2246 which describes TLS
("Transport Layer Security"), which is simply a cleaned up and
standardized version of SSL.

You can only use TLS with an LDAPv3 server. That is because the
standard (RFC 4511) for LDAP and TLS requires that the I<normal> LDAP
connection (ie., on port 389) can be switched on demand from plain text
into a TLS connection. The switching mechanism uses a special extended
LDAP operation, and since these are not legal in LDAPv2, you can only
switch to TLS on an LDAPv3 connection.

So the way you use TLS with LDAPv3 is that you create your normal
LDAPv3 connection using C<Net::LDAP::new()>, and then you perform the
switch using C<Net::LDAP::start_tls()>. The C<start_tls()> method takes
pretty much the same arguments as C<Net::LDAPS::new()>, so check above for
details.

=head2 How does SASL work

SASL is an authentication framework that can be used by a number of
different Internet services, including LDAPv3. Because it is only a
framework, it doesn't provide any way to authenticate by itself; to
actually authenticate to a service you need to use a specific SASL
I<mechanism>. A number of mechanisms are defined, such as CRAM-MD5.

The use of a mechanism like CRAM-MD5 provides a solution to the
password sniffing vulnerability, because these mechanisms typically do
not require the user to send across a secret (eg., a password) in the
clear across the network. Instead, authentication is carried out in a
clever way which avoids this, and so prevents passwords from being
sniffed.

B<Net::LDAP> supports SASL using the B<Authen::SASL> class. Currently the
only B<Authen::SASL> subclasses (ie., SASL mechanism) available are
CRAM-MD5 and EXTERNAL.

Some SASL mechanisms provide a general solution to the sniffing of all
data on the network vulnerability, as they can negotiate confidential
(ie., encrypted) network connections. Note that this is over and above
any SSL or TLS encryption! Unfortunately, perl's B<Authen::SASL> code
cannot negotiate this.

=head1 SEE ALSO

L<Net::LDAP>,
L<Net::LDAPS>,
L<Authen::SASL>

=head1 ACKNOWLEDGEMENTS

Jim Dutton E<lt>jimd@dutton3.it.siu.eduE<gt> provided lots of useful feedback
on the early drafts.

=head1 AUTHOR

Chris Ridd E<lt>chris.ridd@isode.comE<gt>

Please report any bugs, or post any suggestions, to the perl-ldap mailing list
E<lt>perl-ldap@perl.orgE<gt>.

=head1 COPYRIGHT

Copyright (c) 2001-2004 Chris Ridd. All rights reserved. This program is
free software; you can redistribute it and/or modify it under the same
terms as Perl itself.

=cut