Various minor adjustments to code and a WHOLE lot of documentation

[catagits/Catalyst-Plugin-Authentication.git] / lib / Catalyst / Plugin / Authentication / Internals.pod

=head1 NAME

Catalyst::Plugin::Authentication::Internals - All about authentication Stores and Credentials

=head1 INTRODUCTION

L<Catalyst::Plugin::Authentication|Catalyst::Plugin::Authentication> provides
a standard authentication interface to application developers using the
Catalyst framework. It is designed to allow application developers to use
various methods of user storage and credential verification. It is also
designed to provide for minimal change to the application when switching
between different storage and credential verification methods.

While L<Catalyst::Plugin::Authentication|Catalyst::Plugin::Authentication>
provides the interface to the application developer, the actual work of
verifying the credentials and retrieving users is delegated to separate
modules. These modules are called B<Credentials> and storage backends, or
B<Stores>, respectively. For authentication to function there must be at least
one credential and one storage backend. A pairing of a store and a credential
is referred to as a B<Realm>. There may be any number of realms defined for an
application, though most applications will not require more than one or two.

The details of using this module can be found in the
L<Catalyst::Plugin::Authentication|Catalyst::Plugin::Authentication>
documentation.

What follows is an explanation of how the module functions internally and what
is required to implement a credential or a store.

=head1 OVERVIEW

There are two main entry points you need to be aware of when writing a store
or credential module. The first is initialization and the second is during the
actual call to the Catalyst application's authenticate method.

=head2 INITIALIZATION

When the authentication module is loaded, it reads it's configuration to
determine the realms to set up for the application and which realm is to be
the default. For each realm defined in the application's config,
L<Catalyst::Plugin::Authentication|Catalyst::Plugin::Authentication>
instantiates both a new credential object and a new store object. See below
for the details of how credentials and stores are instantiated.

    NOTE: The instances created will remain active throughout the entire 
    lifetime of the application, and so should be relatively lightweight. 
    Care should be taken to ensure that they do not grow, or retain 
    information per request, because they will be involved in each 
    authentication request and could therefore substantially
    hurt memory consumption over time.  

=head2 AUTHENTICATION

When C<$c-E<gt>authenticate()> is called from within an application, the
objects created in the initialization process come into play.
C<$c-E<gt>authenticate()> takes two arguments. The first is a hash reference
containing all the information available about the user. This will be used to
locate the user in the store and verify the user's credentials. The second
argument is the realm to authenticate against. If the second argument is
omitted, the default realm is assumed. 

The main authentication module then locates the credential and store objects
for the realm specified and calls the credential object's C<authenticate()>
method. It provides three arguments, first the application object, or C<$c>,
then a reference to the store object, and finally the hashref provided in the
C<$c-E<gt>authenticate> call. The main authentication module expects the
return value to be a reference to a user object upon successful
authentication. If it receives anything aside from a reference, it is
considered to be an authentication failure. Upon success, the returned user is
marked as authenticated and the application can act accordingly, using
C<$c-E<gt>user> to access the authenticated user, etc.

Astute readers will note that the main
L<Catalyst::Plugin::Authentication|Catalyst::Plugin::Authentication> module
does not interact with the store in any way, save for passing a reference to
it to the credential. This is correct. The credential object is responsible
for obtaining the user from the provided store using information from the
userinfo hashref and/or data obtained during the credential verification
process.

=head1 WRITING A STORE

There are two parts to an authentication store, the backend and the user object.

=head2 STORAGE BACKEND

Writing a storage backend is actually quite simple.  There are only five methods
that must be implemented. They are:

    new()           - instantiates the store object
    find_user()     - locates a user using data contained in the hashref
    for_session()   - prepares a user to be stored in the session
    from_session()  - does any restoration required when obtaining a user from the session
    user_supports() - provides information about what the user object supports

=head3 STORE METHODS

=over 4

=item new( $config, $app )

The C<new()> method is called only once, during the setup process of
L<Catalyst::Plugin::Authentication|Catalyst::Plugin::Authentication>. The
first argument, C<$config>, is a hash reference containing the configuration
information for the store module. The second argument is a reference to the
Catalyst application.

    Note that when new() is called, Catalyst has not yet loaded the various 
    controller and model classes, nor is it definite that other plugins have 
    been loaded, so your new() method must not rely on any of those being 
    present.  If any of this is required for your store to function, you should
    defer that part of initialization until the first method call. 
    
The C<new()> method should return a blessed reference to your store object.

=item find_user( $authinfo, $c ) 

This is the workhorse of any authentication store. It's job is to take the
information provided to it via the C<$authinfo> hashref and locate the user
that matches it. It should return a reference to a user object. A return value
of anything else is considered to mean no user was found that matched the
information provided.

How C<find_user()> accomplishes it's job is entirely up to you, the author, as
is what $authinfo is required to contain.  Many stores will simply use a 
username element in $authinfo to locate the user, but more advanced functionality
is possible and you may bend the $authinfo to your needs.  Be aware, however, that
both Credentials and Stores work with the same $authinfo hash, so take care to
avoid overlapping element names.

Please note that this routine may be called numerous times in various
circumstances, and that a successful match for a user here does B<NOT>
necessarily constitute successful authentication. Your store class should
never assume this and in most cases C<$c> B<should not be modified> by your
store object.

=item for_session( $c, $user )

This method is responsible for preparing a user object for storage in the session.
It should return information that can be placed in the session and later used to 
restore a user object (using the C<from_session()> method).  It should therefore
ensure that whatever information provided can be used by the C<from_session()>
method to locate the unique user being saved.  Note that there is no guarantee
that the same Catalyst instance will receive both the C<for_session()> and 
C<from_session()> calls.  You should take care to provide information that can
be used to restore a user, regardless of the current state of the application. 
A good rule of thumb is that if C<from_session()> can revive the user with the
given information even if the Catalyst application has just started up, you are 
in good shape.

=item from_session( $c, $frozenuser )

This method is called whenever a user is being restored from the session.  
C<$frozenuser> contains the information that was stored in the session for the user.
This will under normal circumstances be the exact data your store returned from 
the previous call to C<for_session()>.  C<from_session()> should return a valid
user object.

=item user_supports( $feature, ...  )

This method allows credentials and other objects to inquire as to what the
underlying user object is capable of. This is pretty-well free-form and the
main purpose is to allow graceful integration with credentials and
applications that may provide advanced functionality based on whether the
underlying user object can do certain things. In most cases you will want to
pass this directly to the underlying user class' C<supports> method. Note that 
this is used as a B<class> method against the user class and therefore must 
be able to function without an instantiated user object.

=back

=head2 USER OBJECT

The user object is an important piece of your store module. It will be the
part of the system that the application developer will interact with most. As
such, the API for the user object is very rigid. All user objects B<MUST>
inherit from
L<Catalyst::Plugin::Authentication::User|Catalyst::Plugin::Authentication::User>.

=head3 USER METHODS

The routines required by the
L<Catalyst::Plugin::Authentication|Catalyst::Plugin::Authentication> plugin
are below. Note that of these, only get_object is strictly required, as the
L<Catalyst::Plugin::Authentication::User|Catalyst::Plugin::Authentication::User>
base class contains reasonable implementations of the rest. If you do choose
to implement only the C<get_object()> routine, please read the base class
documentation so that you fully understand how the other routines will be
implemented for you.

Also, your user object can implement whatever additional methods you require
to provide the functionality you need. So long as the below are implemented,
and you don't overlap the base class' methods with incompatible routines, you
should experience no problems.

=over 4

=item id( )

The C<id()> method should return a unique id (scalar) that can be used to 
retreive this user from the store.  Often this will be provided to the store's
C<find_user()> routine as C<id =E<gt> $user-E<gt>id> so you should ensure that your 
store's C<find_user()> can cope with that. 

=item supports_features( )

This method should return a hashref of 'extra' features supported. This is for
more flexible integration with some Credentials / applications. It is not
required that you support anything, and returning C<undef> is perfectly
acceptable and in most cases what you will do.

=item get( $fieldname )

This method should return the value of the field matching fieldname provided,
or undef if there is no field matching that fieldname. In most cases this will
access the underlying storage mechanism for the user data and return the
information. This is used as a standard method of accessing an authenticated
user's data, and MUST be implemented by all user objects.  

    Note: There is no equivalent 'set' method. Each user class is likely
    to vary greatly in how data must be saved and it is therefore impractical to 
    try to provide a standard way of accomplishing it. When an application 
    developer needs to save data, they should obtain the underlying object / data
    by calling get_object, and work with it directly.
    

=item get_object( )

This method returns the underlying user object. If your user object is backed
by another object class, this method should return that underlying object.
This allows the application developer to obtain an editable object. Generally
speaking this will only be done by developers who know what they are doing and
require advanced functionality which is either unforeseen or inconsistent
across user classes. If your object is not backed by another class, or you
need to provide additional intermediate functionality, it is perfectly
reasonable to return C<$self>.

=back 


... Documentation fairy fell asleep here.  


=head1 MULTIPLE BACKENDS

A key issue to understand about authentication stores is that there are
potentially many of them. Each one is registered into the application, and has
a name.

For most applications, there is only one, and in this framework it is called
'default'.

When you use a plugin, like

    use Catalyst qw/
        Authentication
        Authentication::Store::Foo
    /;

the Store plugins typically only act at setup time. They rarely do more than
check out the configuration, and register e.g. Store::Foo::Backend, and set it
as the default store.

    __PACKAGE__->default_auth_store( $store );

    # the same as

    __PACKAGE__->register_auth_stores( default => $store );

=head1 WORKING WITH USERS

All credential verifiers should accept either a user object, or a user ID.

If a user ID is provided, then they will fetch the user object from the default
store, and check against it.

This should be pretty much DWIM all the time.

When you need multiple authentication backends per application then you must
fetch things yourself. For example:

    my $user = $c->get_auth_store("other_store")->get_user($id);

    $c->login( $user, $supplied_password );

Instead of just:

    $c->login( $id, $supplied_password );

which will go to the default store.

=head1 WRITING A BACKEND

Writing an authentication storage backend is a very simple matter.

The only method you really need to support is C<get_user>.

This method should accept an arbitrary list of parameters (determined by you or
the credential verifyer), and return an object inheriting
L<Catalyst::Plugin::Authentication::User>.

For introspection purposes you can also define the C<user_supports> method. See
below for optional features. This is not necessary, but might be in the future.

=head2 Integrating with Catalyst::Plugin::Session

If your users support sessions, your store should also define the
C<from_session> method. When the user object is saved in the session the
C<for_session> method is called, and that is used as the value in the session
(typically a user id). The store is also saved in the hash. If
C<<$user->store>> returns something registered, that store's name is used. If
not, the user's class is used as if it were a store (and must also support
C<from_session>).

=head2 Optional Features

Each user has the C<supports> method. For example:

    $user->supports(qw/password clear/);

should return a true value if this specific user has a clear text password.

This is on a per user (not necessarily a per store) basis. To make assumptions
about the store as a whole,

    $store->user_supports(qw/password clear/);

is supposed to be the lowest common denominator.

The standardization of these values is to be goverened by the community,
typically defined by the credential verification plugins.

=head2 Stores implying certain credentials

Sometimes a store is agnostic to the credentials (DB storage, for example), but
sometimes it isn't (like an Htpasswd file).

If you are writing a backend that wraps around a module, like
L<Catalyst::Plugin::Authentication::Store::Htpasswd> wraps around
L<Authen::Htpasswd>, it makes sense to delegate the credential checks.

This particular example caused the following "feature" to be added:

    $user->supports(qw/password self_check/);

=head2 Writing a plugin to go with the backend

Typically the backend will do the heavy lifting, by registering a store.

These plugins should look something like this:

    sub setup {
        my $c = shift;

        $c->default_auth_store(
            # a store can be an object or a class
            Catalyst::Plugin::Authentication::Store::Foo::Backend->new(
                ...
            )
        );

        $c->NEXT::setup(@_);
    }