@node Authentication methods
@chapter Authentication methods
@cindex authentication methods

The initial key exchange of the TLS protocol performs authentication
of the peers. In typical scenarios the server is authenticated to
the client, and optionally the client to the server.

While many associate TLS with X.509 certificates and public key
authentication, the protocol supports various authentication methods,
including pre-shared keys, and passwords. In this chapter a description
of the existing authentication methods is provided, as well as some
guidance on which use-cases each method can be used at.

@menu
* Certificate authentication::
* More on certificate authentication::
* Shared-key and anonymous authentication::
* Selecting an appropriate authentication method::
@end menu

@include cha-cert-auth.texi

@include cha-cert-auth2.texi

@include cha-shared-key.texi

@node Selecting an appropriate authentication method
@section Selecting an appropriate authentication method

This section provides some guidance on how to use the available authentication
methods in @acronym{GnuTLS} in various scenarios.

@subsection Two peers with an out-of-band channel

Let's consider two peers need to communicate over an untrusted channel
(the Internet), but have an out-of-band channel available. The latter
channel is considered safe from eavesdropping and message modification and thus
can be used for an initial bootstrapping of the protocol. The options 
available are:
@itemize
@item Pre-shared keys (see @ref{PSK authentication}). The server and a
client communicate a shared randomly generated key over the trusted
channel and use it to negotiate further sessions over the untrusted channel.

@item Passwords (see @ref{SRP authentication}). The client communicates
to the server his username and password of choice and uses it to
negotiate further sessions over the untrusted channel.

@item Public keys (see @ref{Certificate authentication}). The client 
and the server exchange their public keys (or fingerprints of them)
over the trusted channel. 
On future sessions over the untrusted channel they verify the key
being the same (similar to @ref{Verifying a certificate using trust on first use
authentication}).
@end itemize

Provided that the out-of-band channel is trusted all of the above provide
a similar level of protection. An out-of-band channel may be the initial
bootstrapping of a user's PC in a corporate environment, in-person
communication, communication over an alternative network (e.g. the phone
network), etc.

@subsection Two peers without an out-of-band channel

When an out-of-band channel is not available a peer cannot be reliably
authenticated. What can be done, however, is to allow some form of
registration of users connecting for the first time and ensure that their
keys remain the same after that initial connection. This is termed
key continuity or trust on first use (TOFU).

The available option is to use public key authentication (see @ref{Certificate authentication}). 
The client and the server store each other's public keys (or fingerprints of them)
and associate them with their identity.
On future sessions over the untrusted channel they verify the keys
being the same (see @ref{Verifying a certificate using trust on first use
authentication}).

To mitigate the uncertainty of the information exchanged in the first 
connection other channels over the Internet may be used, e.g., @acronym{DNSSEC}
(see @ref{Verifying a certificate using DANE}).

@subsection Two peers and a trusted third party

When a trusted third party is available the most suitable option is to use 
certificate authentication (see @ref{Certificate authentication}). 
The client and the server obtain certificates that associate their identity
and public keys in a reliable way and use them to on the subsequent
communications with each other. Each party verifies the peer's certificate
using the trusted third party's certificate.

While the above is the typical authentication method for servers in the
Internet by using the commercial CAs, the users that act as clients in the
protocol rarely possess such certificates. In that case a hybrid method
can be used where the server is authenticated by the client using the
commercial CAs and the client is authenticated based on some information
the client provided over the initial server-authenticated channel. The
available options are:
@itemize
@item Passwords (see @ref{SRP authentication}). The client communicates
to the server his username and password of choice on the initial
server-authenticated connection and uses it to negotiate further sessions. 
This is possible because the SRP protocol allows for the server to be 
authenticated using a certificate and the client using the
password.

@item Public keys (see @ref{Certificate authentication}). The client 
sends its public key to the server (or a fingerprint of it) over the
initial server-authenticated connection. 
On future sessions the client verifies the server using the third party
certificate and the server verifies that the client's public key remained 
the same (see @ref{Verifying a certificate using trust on first use
authentication}).
@end itemize