
RLM Module Interface (for developers)
-------------------------------------

Section 1 - overview

The fundamental concepts of the rlm interface are module, instance, and
component.

A module is a chunk of code that knows how to deal with a particular kind of
database, or perhaps a collection of similar databases. Examples:

  - rlm_sql contains code for talking to MySQL or Postgres, and for mapping
    RADIUS records onto SQL tables.
  - rlm_unix contains code for making radiusd fit in well on unix-like
    systems, including getpw* authentication and utmp/wtmp-style logging.

An instance specifies the actual location of a collection data that can be
used by a module. Examples:

  - /var/log/radutmp
  - "the MySQL database on bigserver.theisp.com.example"

A module can have up to 4 (soon to be 5? 6?) components which act on RADIUS
requests at different stages. The components are:

  - authorization: check that a user exists, decide on an authentication
    method or proxy realm, and possibly apply some attributes to be returned
    in the reply packet.
  - authentication: verify that the password is correct.
  - preaccounting: decide whether to proxy the request, and possibly add
    attributes that should be included in any logs
  - accounting: record the request in the log
  - checksimul (AVAILABLE AS EXTERNAL PATCH): count the number of active
    sessions for the user
  - postproxyauth (DOES NOT EXIST YET): filter attributes from a reply to a
    proxied request

A module declares which components it supports by putting function pointers
in its "module_t rlm_*" structure.

Section 2 - Module configuration:

The administrator requests the creation of a module instance by adding it
inside the modules{} block in radiusd.conf. The instance definition looks
like this:

  module_name [instance_name] {
    param1 = value1
    param2 = value2
    param3 = value3
    ...
  }

The module_name is used to load the module. To see the names of the available
modules, look for the rlm_*.so files in $installprefix/lib. The module_name
is that, minus the rlm_ and the .so.

instance_name is an identifier for distinguishing multiple instances of the
same module. If you are only loading a module once, you can leave out the
instance_name and it will be assumed to be the same as the module_name.

The parameters inside the module block are passed without interpretation to
the module and generally point to the exact location of a database or enable
optional features of the module. Each module should document what parameters
it accepts and what they do.

For each Access-Request that comes to the server, the authorize{} block is
called. Then one of the authtype{} blocks from authenticate{} is called,
depending on the Auth-Type attribute that was chosen by authorize{}. If
authorize{} set the Proxy-To-Realm attribute, then proxying takes over and
the local authenticate{} phase is skipped.

For each Accounting-Request that comes to the server, the preacct{} block is
called, followed by the accounting{} block. accounting{} is skipped if
preacct{} sets Proxy-To-Realm.

For an explanation of what "calling" a config block means, see
README.failover.

Section 3 - The lifetime of a module

When the server is starting up, or reinitializing itself as a result of a
SIGHUP, it reads the modules{} section. Each configured module will be loaded
and its init() method will be called.

  int init(void)

The init() method should take care of
any setup that is not tied to a specific instance. It will only be called
once, even if there are multiple instances configured.

For each configured instance, after the init() method, the instantiate()
method is called. It is given a handle to the configuration block holding its
parameters, which it can access with cf_section_parse().

  int instantiate(CONF_SECTION *cs, void **instance)

The instantiate() function should look up options in the config section, and
open whatever files or network connections are necessary for the module to do
its job. It also should create a structure holding all of the persistent
variables that are particular to this instance (open file descriptors,
configured pathnames, etc.) and store a pointer to it in *instance. That
void * becomes a handle (some would call it a "cookie") representing this
instance. The instance handle is passed as a parameter in all subsequent
calls to the module's methods, so they can determine which database they are
supposed to act on.

The authorize(), authenticate(), preaccounting(), and accounting() functions
are all called the same way:

  int authorize(void *instance, REQUEST *request)
  int authenticate(void *instance, REQUEST *request)
  int preaccounting(void *instance, REQUEST *request)
  int accounting(void *instance, REQUEST *request)

they each receive the instance handle and the request, and are expected to
act on the request using the database pointed to by the instance handle
(which was set by the instantiate() function).

When the server is being shut down (as the first part of SIGHUP for example)
detach() is called for each module instance.

  int detach(void *instance)

The detach() method should release whatever resources were allocated by the
instantiate() method.

After all instances are detached, the destroy() method is called.

  int destroy(void)

It should release resources that were acquired by the init() method.

--Alan Curry <pacman@world.std.com>
