See http://www.kame.net/newsletter/20001119b/

The gss-api authentication mechanism implementation for racoon was
based on the ietf draft draft-ietf-ipsec-isakmp-gss-auth-06.txt.

The implementation uses the Heimdal gss-api library, i.e. gss-api
on top of Kerberos 5. The Heimdal gss-api library had to be modified
to meet the requirements of using gss-api in a daemon. More specifically,
the gss_acquire_cred() call did not work for other cases than
GSS_C_NO_CREDENTIAL ("use default creds"). Daemons are often started
as root, and have no Kerberos 5 credentials, so racoon explicitly
needs to acquire its credentials. The usual method (already used
by login authentication daemons) in these situations is to add
a set of special credentials to be used. For example, authentication
by daemons concerned with login credentials, uses 'host/fqdn' as
its credential, where fqdn is the hostname on the interface that
is being used. These special credentials need to be extracted into
a local keytab from the kdc. The default value used in racoon
is 'ike/fqdn', but it can be overridden in the racoon config file.

The modification to the Heimdal gss-api library implements the
mechanism above. If a credential other than GSS_C_NO_CREDENTIAL
is specified to gss_acquire_cred(), it first looks in the default
credential cache if it its principal matches the desired credential.
If not, it extracts it from the default keytab file, and stores
it in a memory-based credential cache, part of the gss credential
structure.



The modifcations to racoon itself are as follows:

	* The racoon.conf config file accepts a new keyword, "gssapi_id",
	  to be used inside a proposal specification. It specifies
	  a string (a Kerberos 5 principal in this case), specifying the
	  credential that racoon will try to acquire. The default value
	  is 'ike/fqdn', where fqdn is the hostname for the interface
	  being used for the exchange. If the id is not specified, no
	  GSS endpoint attribute will be specified in the first SA sent.
	  However, if the initiator does specify a GSS endpoint attribute,
	  racoon will always respond with its own GSS endpoint name
	  in the SA (the default one if not specified by this option).

	* The racoon.conf file accepts "gssapi_krb" as authentication
	  method inside a proposal specification. The number used
	  for this method is 65001, which is a temporary number as
	  specified in the draft.

	* The cftoken.l and cfparse.y source files were modified to
	  pick up the configuration options. The original sources
	  stored algorithms in bitmask, which unfortunately meant
	  that the maximum value was 32, clearly not enough for 65001.
	  After consulting with the author (sakane@kame.net), it turned
	  out that method was a leftover, and no longer needed. I replaced
	  it with plain integers.

	* The gss-api specific code was concentrated as much as possible
	  in gssapi.c and gssapi.h. The code to call functions defined
	  in these files is conditional on HAVE_GSSAPI, except for the
	  config scan code. Specifying this flag on the compiler commandline
	  is conditional on the --enable-gssapi option to the configure
	  script.

	* Racoon seems to want to send accepted SA proposals back to
	  the initiator in a verbatim fashion, leaving no room to
	  insert the (variable-length) GSS endpoint name attribute.
	  I worked around this by re-assembling the extracted SA
	  into a new SA if the gssapi_krb method is used, and the
	  initiator sent the name attribute. This scheme should
	  possibly be re-examined by the racoon maintainers, storing
	  the SAs (the transformations, to be more precise) in a different
	  fashion to allow for variable-length attributes to be
	  re-inserted would be a good change, but I considered it to be
	  beyond the scope of this project.

	* The various state functions for aggressive and main mode
	  (in isakmp_agg.c and isakmp_ident.c respectively) were
	  changed to conditionally change their behavior if the
	  gssapi_krb method is specified.


This implementation tried to follow the specification in the ietf draft
as close as possible. However, it has not been tested against other
IKE daemon implementations. The only other one I know of is Windows 2000,
and it has some caveats. I attempted to be Windows 2000 compatible.
Should racoon be tried against Windows 2000, the gssapi_id option in
the config file must be used, as Windows 2000 expects the GSS endpoint
name to be sent at all times. I have my doubts as to the W2K compatibility,
because the spec describes the GSS endpoint name sent by W2K as
an unicode string 'xxx@domain', which doesn't seem to match the
required standard for gss-api + kerberos 5 (i.e. I am fairly certain
that such a string will be rejected by the Heimdal gss-api library, as it
is not a valid Kerberos 5 principal).

With the Heimdal gss-api implementation, the gssapi_krb authentication
method will only work in main mode. Aggressive mode does not allow
for the extra round-trips needed by gss_init_sec_context and
gss_accept_sec_context when mutual authentication is requested.
The draft specifies that the a fallback should be done to main mode,
through the return of INVALID-EXCHANGE-TYPE if it turns out that
the gss-api mechanisms needs more roundtrips. This is implemented.
Unfortunately, racoon does not seem to properly fall back to
its next mode, and this is not specific to the gssapi_krb method.
So, to avoid problems, only specify main mode in the config file.


	-- Frank van der Linden <fvdl@wasabisystems.com>

HOW-TO use plainrsa auth, contributed by Simon Chang <simonychang@gmail.com>

Before you begin, you should understand that the RSA authentication
mechanism hinges upon the idea of a split cryptographic key:  one used
by the public, the other readable only to you.  Any data that is
encrypted by a public key can be decrypted only by the corresponding
private key, so that the private key user can be assured that the
content of the transmission has not been examined by unauthorized
parties.  Similarly, any data encrypted by the private key can be
decrypted by the public key so that the public knows that this
transmission came from this user and nobody else (this idea is called
non-repudiation).  Also, the longer the key length, the more difficult
it would be for potential attacker to conduct brute-force discovery of
the keys.  So, what all this means for the security administrator is
that the setup needs a pair of reasonably long keys for each host that
wishes to authenticate in this manner.

With this in mind, it should be relatively straightforward to set up
RSA authentication.  For the purpose of this document, we assume that
we are setting up RSA authentication between two networked hosts
called Boston and Chicago.  Unless otherwise noted, all steps should
be performed on both hosts with corresponding key names.  Here are the
steps:

1)  Included in each default installation of ipsec-tools is a binary
called plainrsa-gen.  This executable is used to generate a pair of
RSA keys for the host.  There are only two parameters that you should
be concerned about: -b, which sets the number of bits for the keys,
and -f, which specifies the output file for plainrsa-gen to send the
results.  On an ordinary Pentium-II with 128 MB of RAM, it takes only
seconds to generate keys that are 2048 bits long, and only slightly
longer to generate 4096-bit keys.  Either key length should be
sufficient; any longer key length actually reduces performance and
does not increase security significantly.  You should therefore run it
as:

	plainrsa-gen -b 2048 -f /var/tmp/boston.keys

2)  When the process completes, you should have a text file that
includes both public and private keys.  GUARD THIS FILE CAREFULLY,
because once a private key is compromised it is no longer any good,
and you must generate a new pair from scratch.  Reading the file
itself, you should see several very long lines of alphanumeric data.
The only line you should be concerned with is the line towards the top
of the output file that begins with "# pubkey=0sAQPAmBdT/" or
something to that effect.  This line is your public key, which should
be made available to the other host that you are setting up.  Copy
this line to a separate file called "boston.pub" and change the
beginning of the line so that it reads ": PUB 0sAQPAmBdT/".
Alternatively, you can also grab the first line of the boston.keys
file and uncomment the line so that it reads the same as above.  Now
rename the file you generated initially to "boston.priv".

3)  You should now have two files, boston.priv and boston.pub
(chicago.priv and chicago.pub on Chicago).  The first file contains
your private key and the second file your public key.  Next you should
find a way to get the public key over to the other host involved.
Boston should have (1) its own key pair, and (2) Chicago's public key
ONLY.  Do not copy Chicago's private key over to Boston, because (a)
it is not necessary, and (b) you would now have two potential places
for losing control of your private key.

4)  You should now configure the racoon.conf configuration file for
each host to (a) turn on RSA authentication, and (b) designate each
host's private key and the remote host(s)'s public key(s).  Take all
your keys and place it in one directory and use the global directive
"path certificate" to specify the location of the keys.  This step is
especially important if you are running racoon with privilege
separation, because if racoon cannot find the keys inside the
directory you have just specified it will fail the authentication
process.  So, write the directive like the following:

	path certificate "/etc/racoon";

Next, you need to specify the host's own private key and the public
keys of all the remote peers involved. For your local private key and
remote public key(s), you should use the following directives:

	certificate_type plain_rsa "/etc/racoon/boston.priv";
	peers_certfile plain_rsa "/etc/racoon/chicago.pub";

Notice the option "plain_rsa" for both directives.

Finally, under the "proposal" statement section, you should specify
the "rsasig" option for "authentication_method".

5)  You have finished configuring the host for RSA authentication.
Now use racoonctl to reload the configuration or simply restart the
machine and you should be all set.

TROUBLESHOOTING

In the event that the hosts fail to communicate, first go back to the
instructions above and make sure that:

1)  You have placed all the keys in the directory that is specified by
the "path certificate" directive.  Keep in mind that privilege
separation will force racoon to look into that directory and nowhere
else.
2)  You have specified correctly the host's own private key and the
remote peer's public key.
3)  You have specified the "rsasig" method for authentication in the
proposal statement.

If you run into any further problems, you should try to use "racoon
-v" to debug the setup, and send a copy of the debug messages to the
mailing list so that we can help you determine what the problem is.

Last modified: $Date: 2006/12/10 05:51:14 $