$user->supports("foo") will also return true for User::Hash if there is such a key

[catagits/Catalyst-Plugin-Authentication.git] / notes.pod

=pod

=head1 CORE CONCEPTS

=item authentication

Authentication means ensuring the user is who they claim to be.	

=item authorization

Once a user's identity is known and authenticated, we can decide what they are
authorized to do.

=head2 Authentication

The process of authenticating the user is, almost always, asking the user to
prove they know a secret that only the real user would know.

Ways of proving you know a secret are:

=over 4

=item cleartext password

Just tell the secret to the other side to prove you know it.

=item challange/response

Prove you know the secret without revealing it, and without the possibility of
an attacker gaining access even if they get a copy of the data sent from the
client.

=over 4

=item 1

The server sends some random data, unique for each request

=item 2

The client combines the random data with the secret, and digests that

=item 3

The server combines the data with the password as well, and makes sure the
digest is the same as what the user sent.

=back

Can possibly be implemented optionally using javascript on the client side if
it's enabled.

=item public key cryptography

...

=back

=head2 Authorization

=over 4

=item Role based classification 

Each user plays in several roles. Restricted actions can then check that the
user belongs to a certain role (for example, for the 'delete' action you might
want to ensure that a user is an 'editor').

Since users are allowed to be in an arbitrary amount of roles the logic is
simple and effective.

=item ACLs

Cascading on top of roles, and more tightly coupled with catalyst is the
possibility of access control lists in two namespaces:

=over 4

=item *

The perl package namespace

=item *

The mapped URI space

=back

For example, you could say that the controller MyApp::C::AdminPanel requires
the role 'admin'.

This should resemble the way apache can provide access control for
C<< <Location ...> >> and C<< <Directory ...> >>.

=back

=head1 ASPECTS OF IMPLEMENTATION

=head2 Authentication

Authentication needs to take care of two things:

=over 4

=item *

Storage of the user data

=item *

Authentication of credentials against storage

=back

Ideally the two are decoupled, with the following call chain relationship:

When application asks the authentication plugin to verify credentials, the
credential verification plugin asks storage retrieve the user based on the user
ID (the credential plugin knows what part of the credential is a user ID).

A storage must return an object representing the user, whose API is irrelevant
here.

It is up to the user to synchronize the user info storage with the credential
plugin, so that the value retrieved from storage is the value the credential
plugin expects.

For example, the user/password credential plugin will look something like this:

	sub authen_login {
		my ( $c, %opts ) = @_;
		my $user = $opts{user} || $c->auth_get_user( %opts ); # note that
		# credential plugin methods should accept a user object if they are
		# provided with one

		if ( $user->password eq $opts->{password} ) {
			...
		}
	}

A situation where login/password authentication would have to be changed is if
the credential system uses /etc/passwd storage, for example. Since /etc/passwd
contains hashes of passwords, we can't compare the password directly. The
credential system must know to take care of this.

The login/password plugin should probably be a bit more lax, making checks on
$user to see what the user supports:

	if (my $hash = eval { $user->hashed_password }) {
		my $digest = Digest->new( $user->hash_algorithm );
		...
		$digest->digest eq $hash; # yes, this is binay, not hex or whatever
	} elsif (my $crypt = eval { $user->crypted_password }) {
		return crypt( $password, $crypt ) eq $crypt;
	} elsif (my $clear = eval { $user->password }) {
		return $passwd eq $clear;
	} else {
		Catalyst::Exception->throw("password storage scheme possibilities exhausted");
	}

Or less optimistically

	if ( $user->supports(password => "hashed") ) {
		my $hash = $user->hashed_password;
		...
	}

C<supports> in the user base class should look like this:

	sub supports {
		my ( $self, @path ) = @_;

		my $cursor = $self->_supports; # this is a nested hash

		while (@path) {
			ref $cursor or return undef;
			$cursor = $cursor->{ shift @path };
		}

		return $cursor && !ref($cursor);
	}

Which just goes through a nested hash, populated by the implementation class.

The structure of this hash should be converged by the community, and should
look something like this for a plugin that supports everything I can think of:

	{
		password => {
			hash => 1,
			crypt => 1,
			clear => 1,
		},
		key {
			gpg => 1,
			ssh => 1,
			...
		},
		secureid => 1,
		kerberos => 1, # maybe this needs to be deeper? i dunno
	},


In the event that the storage backend implies a certain format, the credential
verification and storage plugins should ship together, for convenience.

=head3 Testing for authentication

Once authenticated the authentication plugin sets the C<user> accessor in the
context object to contain the user object.

=head3 Stores are just models

To make things more flexible, stores are really just models.

To make things easy a "default" auth storage can be [automatically] selected
for the user for the more common situation of one auth driver per app:

	sub auth_get_user {
		my ( $c, %opts ) = @_;

		( $opts->{store} || $c->find_auth_store )->auth_get_user( %opts );
	}

	sub find_auth_store {
		my $c = shift;

		$c->config->{auth}{default_store} || ...;
	}

Ofcourse, %opts should not be just delegated blindly in the production
versions. Perhaps even L<Params::Validate> is in order.

=head3 Integration with Catalyst::Plugin::Session

If a session plugin is loaded (C<< $c->isa("Catalyst::Plugin::Session) >>), it
should also store the user ID in C<< $c->session->{user} >> unless
C<< $c->config->{authentication}{use_session} >> is a false value (true by
default).

=head2 Authorization

Authorization uses the same storage backend as authentication - that's why they
are decoupled.

Like the credential verification part, the authorization system simply assumes
that the necesseary information is in the object representing the user.

The API required should be strictly documented, in order to minimize confusion.

For example, the role authorization plugin may require a C<roles> method:

	use Quantum::Superpositions qw/all any/;

	sub authz_check_roles {
		my ( $c, @required_roles ) = @_;

		all(@required_roles) == $c->user->roles;
	}

Which must be arranged by the authentication storage plugin's user class.

In order to facilitate this all storage plugins, either trivial ones, should
use factory methods for generating the user object. This allows the user to
override the method, replacing the user class with their own, extended class
that provides the necessary glue.

=head1 DESIRED PLUGINS

=head2 Credential Verification

=over 4

=item login/password

Requires a C<password> method from the user, checks for equality

=item unix passwd

Requires a C<crypted_passwd> method from the user, checks for equality after crypt

=item challenge response

pretends to be a login/password compatible plugin, and will fall back to it,
unless the client can use javascript to create a digest of the password and a
seed instead of sending the actual password.

=item http auth

Send 401 status code, read back authentication headers

=item delegated

a passive authentication plugin that checks whether the user logged in using
the engine's backend (e.g. apache) and provides a handle object representing
that user that is not tied to any storage.

=head2 Storage

=over 4

=item DBIC/CDBI

User table, contains arbitrary fields and relationships.

This should simply provide the interface for storage plugins based on a table
in config.

=item LDAP

FIXME I'm not quite sure how LDAP looks from an OO perspective.

=item htpasswd

Implies credential verification for the various types of passwords htpasswd
supports.

Can use extra info field for additional interface glue. For example:

	sub roles {
		my $self = shift;
		split(",", $self->info);
	}

=item passwd

A subset of htpasswd, can use comment field for extended info like htpasswd.

=item perl hash of objects

a hash of user object keyed by user ID can prove very convenient for rapid
development.

=back

=head2 Authorization

=over 4

=item roles

=item ACLs

A complex interface that layers on roles and automatically checks role
membership for certain areas of an application based on a perl data structure.

=back

=head1 GUIDELINES

Avoid new formats - there are enough standard formats out there that we can
use. No need for XML, YAML, etc authentication support in the core of these plugins.

These plugins can be supplemented by ::Simple namespaces that have the extra sugar.

This is one of the reasons Authentication::Simple sucked so bad - it's file
parser was half baked, it was very inefficient, and about 10x as many lines of
code it could have been if the user simply had to put a perl hash of arrays in
the config.


=head1 TODO

=over 4

=item Look into L<Apache::AuthCookie>

 < kgftr|konobi> nothingmuch: Apache::AuthCookie has a really nice way of
 dealing with multiple types of auth (like roles, groups, etc)

=back


=head1 API tree

This tree proposes a list the classes in the authentication/authorization
plugins, as well as their methods.

It's purpose is to help synchronize the authen/authz efforts, and is not to be
considered a formal spec.

	* Authentication (umbrella plugin)
		- set_authenticated( $user ) # set $c->user, and $c->session->{user_id} = $user->id
		- logout # delete $c->user, delete $c->session->{user_id}
		- user # currently logged in user, if any
		- prepare (extended) # checks session to see if a user should be logged in
		* User
			- id (the value that can be used to get the user with)
			- supports
			  method name, or hierarchy
			  The methods listed under a hierarchy mean the methods you can call
			  if $user->supports( qw/password clear/ ) then you can $user->password safely
				* password
					* clear
						- password
					* hashed
						- hashed_password
						- hash_algorithm
					* crypted
						- crypted_password
				* roles
					- roles
				* ...
		* Store (plugin)
		  the storage plugin is a thin wrapper around a store model, that
		  provides a notion of a default store for easy dwimming.
		  it's two methods are just dump delegations to $c->config->{authentication}{store}
			- get_user( $id )
			- user_supports( $id )
			* DBIC
			  generates an object, puts it in $c->{authetnciation}{store}
			  this object inherits the model class but is not really a model,
			  instead it delegates to the real DBIC model under it
			* ... # each store impl has a dual life as a plugin
		* Store::...::Model
		  provides the actual storage model glue. For example, in DBIC get_user
		  will do a retrieve on the user table.
			- get_user( $id ) # $id is username, or other credential a user might type in
			- user_supports
			  like calling ->supports on the user, but applies to all users
			  if one user supports X but another doesn't, the global check will say they both don't
			* Minimal
			  A nested hash in $c->config->{authentication}{entries}
			    {
			      user_id => $user_obj || $hash_ref
				}
			  Hash ref will auto instantiate into a user obj that provides a default
			  implementation of supports and some accessors, based on the keys
			* DBIC
			  storage dependent
				- roles # has_many rel
				- password # column
				- crypted_password # column
				- hashed_password # column
				- hash_algorithm # column
				- ... # columns
			* UNIX (Unix::PasswdFile, Unix::GroupFile)
			  provides simple role based authentication
			  implies the Password credential checker
				- uid
				- name
				- crypted_password
				- gid
				- comment
				- home
				- shell
				- groups/roles # aliases
				- add_group/add_role
				  uses group password to temporary add a group/role (newgrp)
				  stores the added groups in the session data
			* Htpasswd (Apache::Htpasswd)
			  has optional role based authentication in the info field
			  ($c->config->{authentication}{htpasswd}{role_delimiter} defaults to ',')
			  implies the Password credential checker
				- name
				- password # htpasswd -p
				- crypted_password # htpasswd -d
				- hashed_password # htpasswd -s, htpasswd -m
				- hash_algorithm
		* Credential (plugin)
		  all $user args can be either user objects, or user ids that will be queried from
		  $c->config->{authentication}{store}
			* Password
				- login( $user, $password )
				  tries hashed password, crypted password, and password based on caps of user obj
			* CRAM (requires session)
			  challange is the session ID
			  response is hash( challange + ( password | crypted_password | hashed_password ) )
			  possibly backed by javascript on the client side
				- authenticate_cram( $user, $response )
				  hashes $c->sessionid and password variant based on config
			* SRP (probably useless, but at least take a look - SSL uses it)
			  possibly backed by javascript on the client side
			* HTTP Auth  (like Catalyst::Plugin::Authentication::CDBI::Basic does)
			  subclasses Credential::Password, and just calls login automatically from prepare
			  takes a value to $c->forward to in case the user isn't authenticated, to generate 401
			* Apache
			  logs in a user without a store, by checking apache's authentication data
			* Net (chansen)
			  all of these are Password subclasses
				* FTP
				* SMTP
				* POP3 
			* Secure ID
			* SSH (chansen)
		* Store/Credential hybrid
			* LDAP
			  storage dependent... querying the database is tied with credential checking
				- roles (memberOf)
				- ... (attributes)
			* RADIUS (chansen)
			* PAM (chansen)
			* Service
				* OpenID (chansen)
				* Passport (hah!)
				* TypeKey
				* Sxip
			* SAML
			* LID
			* SMB (chansen)
			* Kerberos (chansen)
			* SASL (Authen::SASL)
	* Authorization (plugin)
		* Roles
		  checks a user's membership in roles
		  uses has_roles( @roles ) if the user class has that method (assumed to be more efficient)
		  uses roles and compares on it's own otherwise
			- check_user_roles( [ $user ], @roles )
			  returns boolean, whether $c->user (or supplied user) has specified roles
			- assert_user_roles( [ $user ] @roles )
			  throws an exception if $c->user (or supplied user) doesn't have the specified roles
		* ACL
		  allows to add ACL rules to sections of the catalyst app
		  ACLs apply to perl namespace (MyApp::C::Foo "contains" MyApp::C::Foo::Bar, etc)
		  or they can apply to URL space (/foo contains /bar)
		  implies $c->user
		  takes a value to $c->forward to in case the user isn't authorized (error page)
			- restrict_access ( $arg_like_forward_takes, @roles ) # roles to allow
			- deny_access ( $arg_like_forward_takes, @roles ) # roles to disallow
		* SAML
	* Login Form (plugin)
	  this is the plugin that will be used for CRAM stuff in javascript
	  http://pajhome.org.uk/crypt/ - bsd license hashes in js
	  http://pajhome.org.uk/crypt/md5/chaplogin.html
	  http://perl-md5-login.sourceforge.net/
	  it's just for convenience
		- login_form( [ $store ] )
		  uses user_supports and isa checks on credentials to create a form
		  that makes sense for the store provided if no store is provided uses
		  $c->config->{authenticion}{store}
		- process_login
		  using the same info about stores/credentials, will pick up the data
		  used to generate a form, and find out what form data to use to login
		- prepare
		  if $c->config->{authentication}{auto_login} is on will call process_login automatically