Update of PLUGINS documentation

[ppp.git] / PLUGINS

Starting with version 2.3.10, pppd includes support for `plugins' -
pieces of code which can be loaded into pppd at runtime and which can
affect its behaviour in various ways.  The idea of plugins is to
provide a way for people to customize the behaviour of pppd without
having to either apply local patches to each version or get their
patches accepted into the standard distribution.

A plugin is a standard shared library object, typically with a name
ending in .so.  They are loaded using the standard dlopen() library
call, so plugins are only supported on systems which support shared
libraries and the dlopen call.  At present pppd is compiled with
plugin support only under Linux and Solaris.

Plugins are loaded into pppd using the `plugin' option, which takes
one argument, the name of a shared object file.  The plugin option is
a privileged option.  If the name given does not contain a slash, pppd
will look in the /usr/lib/pppd/<version> directory for the file, where
<version> is the version number of pppd, for example, 2.4.2.  I
suggest that you either give the full path name of the shared object
file or just the base name; if you don't, it may be possible for
unscrupulous users to substitute another shared object file for the
one you mean to load, e.g. by setting the LD_LIBRARY_PATH variable.

Plugins are usually written in C and compiled and linked to a shared
object file in the appropriate manner for your platform.  Using gcc
under Linux, a plugin called `xyz' could be compiled and linked with
the following commands:

        gcc -c -O xyz.c
        gcc -shared -o xyz.so xyz.o

There are some example plugins in the pppd/plugins directory in the
ppp distribution.  Currently there is one example, minconn.c, which
implements a `minconnect' option, which specifies a minimum connect
time before the idle timeout applies.

Plugins can access global variables within pppd, so it is useful for
them to #include "pppd.h" from the pppd source directory. Other
header files can be included such as chap.h, mppe.h, and upap.h as
needed per project.

Every plugin must contain a global procedure called `plugin_init'.
This procedure will get called (with no arguments) immediately after
the plugin is loaded.  Every plugin should also contain a variable
called pppd_version declared as follows:

char pppd_version[] = PPPD_VERSION;

If this declaration is included, pppd will not load the module if its
version number differs from that compiled into the plugin binary.

Plugins can affect the behaviour of pppd in at least four ways:

1. They can add extra options which pppd will then recognize.  This is
   done by calling the ppp_add_options() procedure with a pointer to an
   array of option_t structures.  The last entry in the array must
   have its name field set to NULL.

2. Pppd contains `hook' variables which are procedure pointers.  If a
   given hook is not NULL, pppd will call the procedure it points to
   at the appropriate point in its processing.  The plugin can set any
   of these hooks to point to its own procedures.  See below for a
   description of the hooks which are currently implemented.

3. Plugin code can call any global procedures and access any global
   variables in pppd.

4. Plugins can register procedures to be called when particular events
   occur, using the `notifier' mechanism in pppd.  The differences
   between hooks and notifiers are that a hook will only call one
   function, whereas a notifier can call an arbitrary number, and that
   a hook usually returns some value to pppd, whereas a notifier
   function returns nothing.

Here is a list of the currently implemented hooks in pppd.


int (*idle_time_hook)(struct ppp_idle *idlep);

The idle_time_hook is called when the link first comes up (i.e. when
the first network protocol comes up) and at intervals thereafter.  On
the first call, the idlep parameter is NULL, and the return value is
the number of seconds before pppd should check the link activity, or 0
if there is to be no idle timeout.

On subsequent calls, idlep points to a structure giving the number of
seconds since the last packets were sent and received.  If the return
value is > 0, pppd will wait that many seconds before checking again.
If it is <= 0, that indicates that the link should be terminated due
to lack of activity.


int (*holdoff_hook)(void);

The holdoff_hook is called when an attempt to bring up the link fails,
or the link is terminated, and the persist or demand option was used.
It returns the number of seconds that pppd should wait before trying
to reestablish the link (0 means immediately).


int (*pap_check_hook)(void);
int (*pap_passwd_hook)(char *user, char *passwd);
int (*pap_auth_hook)(char *user, char *passwd, char **msgp,
                     struct wordlist **paddrs,
                     struct wordlist **popts);
void (*pap_logout_hook)(void);

These hooks are designed to allow a plugin to replace the normal PAP
password processing in pppd with something different (e.g. contacting
an external server).

The pap_check_hook is called to check whether there is any possibility
that the peer could authenticate itself to us.  If it returns 1, pppd
will ask the peer to authenticate itself.  If it returns 0, pppd will
not ask the peer to authenticate itself (but if authentication is
required, pppd may exit, or terminate the link before network protocol
negotiation).  If it returns -1, pppd will look in the pap-secrets
file as it would normally.

The pap_passwd_hook is called to determine what username and password
pppd should use in authenticating itself to the peer with PAP.  The
user string will already be initialized, by the `user' option, the
`name' option, or from the hostname, but can be changed if necessary.
MAXNAMELEN bytes of space are available at *user, and MAXSECRETLEN
bytes of space at *passwd.  If this hook returns 0, pppd will use the
values at *user and *passwd; if it returns -1, pppd will look in the
pap-secrets file, or use the value from the +ua or password option, as
it would normally.

The pap_auth_hook is called to determine whether the username and
password supplied by the peer are valid.  user and passwd point to
null-terminated strings containing the username and password supplied
by the peer, with non-printable characters converted to a printable
form.  The pap_auth_hook function should set msg to a string to be
returned to the peer and return 1 if the username/password was valid
and 0 if not.  If the hook returns -1, pppd will look in the
pap-secrets file as usual.

If the username/password was valid, the hook can set *paddrs to point
to a wordlist containing the IP address(es) which the peer is
permitted to use, formatted as in the pap-secrets file.  It can also
set *popts to a wordlist containing any extra options for this user
which pppd should apply at this point.

The pap_logout_hook is called when the link is terminated, instead of
pppd's internal `plogout' function.  It can be used for accounting
purposes.  This hook is deprecated and will be replaced by a notifier.


int (*chap_check_hook)(void);
int (*chap_passwd_hook)(char *user, char *passwd);
int (*chap_verify_hook)(char *name, char *ourname, int id,
                        struct chap_digest_type *digest,
                        unsigned char *challenge, unsigned char *response,
                        char *message, int message_space)

These hooks are designed to allow a plugin to replace the normal CHAP
password processing in pppd with something different (e.g. contacting
an external server).

The chap_check_hook is called to check whether there is any possibility
that the peer could authenticate itself to us.  If it returns 1, pppd
will ask the peer to authenticate itself.  If it returns 0, pppd will
not ask the peer to authenticate itself (but if authentication is
required, pppd may exit, or terminate the link before network protocol
negotiation).  If it returns -1, pppd will look in the chap-secrets
file as it would normally.

The chap_passwd_hook is called to determine what password
pppd should use in authenticating itself to the peer with CHAP.  The
user string will already be initialized, by the `user' option, the
`name' option, or from the hostname, but can be changed if necessary.
This hook is called only if pppd is a client, not if it is a server.

MAXSECRETLEN bytes of space are available at *passwd.  If this hook
returns 0, pppd will use the value *passwd; if it returns -1, pppd
will fail to authenticate.

The chap_verify_hook is called to determine whether the peer's
response to our CHAP challenge is valid -- it should return 1 if valid
or 0 if not.  The parameters are:

* name points to a null-terminated string containing the username
  supplied by the peer, or the remote name specified with the
  "remotename" option.
* ourname points to a null-terminated string containing the name of
  the local machine (the hostname, or the name specified with the
  "name" option).
* id is the value of the id field from the challenge.
* digest points to a chap_digest_type struct, which contains an
  identifier for the type of digest in use plus function pointers for
  functions for dealing with digests of that type.
* challenge points to the challenge as a counted string (length byte
  followed by the actual challenge bytes).
* response points to the response as a counted string.
* message points to an area of message_space bytes in which to store
  any message that should be returned to the peer.


int (*null_auth_hook)(struct wordlist **paddrs,
                      struct wordlist **popts);

This hook allows a plugin to determine what the policy should be if
the peer refuses to authenticate when it is requested to.  If the
return value is 0, the link will be terminated; if it is 1, the
connection is allowed to proceed, and in this case *paddrs and *popts
can be set as for pap_auth_hook, to specify what IP addresses are
permitted and any extra options to be applied.  If the return value is
-1, pppd will look in the pap-secrets file as usual.


void (*ip_choose_hook)(u_int32_t *addrp);

This hook is called at the beginning of IPCP negotiation.  It gives a
plugin the opportunity to set the IP address for the peer; the address
should be stored in *addrp.  If nothing is stored in *addrp, pppd will
determine the peer's address in the usual manner.


int (*allowed_address_hook)(u_int32_t addr)

This hook is called to see if a peer is allowed to use the specified
address.  If the hook returns 1, the address is accepted.  If it returns
0, the address is rejected.  If it returns -1, the address is verified
in the normal away against the appropriate options and secrets files.


void (*snoop_recv_hook)(unsigned char *p, int len)
void (*snoop_send_hook)(unsigned char *p, int len)

These hooks are called whenever pppd receives or sends a packet.  The
packet is in p; its length is len.  This allows plugins to "snoop in"
on the pppd conversation.  The hooks may prove useful in implmenting
L2TP.


void (*multilink_join_hook)();

This is called whenever a new link completes LCP negotiation and joins
the bundle, if we are doing multilink.


A plugin registers itself with a notifier by declaring a procedure of
the form:

void (ppp_notify_fn)(void *opaque, int arg);

and then registering the procedure with the appropriate notifier with
a call of the form

        ppp_add_notify(ppp_notify_t, ppp_notify_fn, opaque);

The ppp_notify_t is an enumerated type that describes which notifier
to attach the function to. Example: NF_EXIT, NF_SIGNALED, NF_IP_UP

The `opaque' parameter in the add_notifier call will be passed to
my_notify_proc every time it is called.  The `arg' parameter to
my_notify_proc depends on the notifier.

A notify procedure can be removed from the list for a notifier with a
call of the form

        ppp_del_notify(ppp_notify_t, ppp_notify_fn, opaque);

Here is a list of the currently-implemented notifiers in pppd.

* NF_PID_CHANGE.  This notifier is called in the parent when pppd has
  forked and the child is continuing pppd's processing, i.e. when pppd
  detaches from its controlling terminal.  The argument is the pid of
  the child.

* NF_PHASE_CHANGE.  This is called when pppd moves from one phase of
  operation to another.  The argument is the new phase number.

* NF_EXIT.  This is called just before pppd exits.  The argument is
  the status with which pppd will exit (i.e. the argument to exit()).

* NF_SIGNALED.  This is called when a signal is received, from within
  the signal handler.  The argument is the signal number.

* NF_IP_UP.  This is called when IPCP has come up.

* NF_IP_DOWN.  This is called when IPCP goes down.

* NF_IPV6_UP.  This is called when IP6CP has come up.

* NF_IPV6_DOWN.  This is called when IP6CP goes down.

* NF_AUTH_UP.  This is called when the peer has successfully
  authenticated itself.

* NF_LINK_DOWN.  This is called when the link goes down.

* NF_FORK.  Called for each time pppd exists as a new process (child). 


Regarding MPPE keys and key-material for 2.5.0 release

Sometimes it is necessary for a plugin to access details related to
the authentication process. The NF_AUTH_UP callback notifier (client only)
allows a plugin to inspect e.g. key details after authentication has been
completed, but before the key material is cleared from memory for security
reasons.

There are in particularly 3 functions that allow one to inspect these
keys:

* bool mppe_keys_isset()
* int mppe_get_recv_key(unsigned char *key, int length)
* int mppe_get_send_key(unsigned char *key, int length)

The first function indicates whether or not the key material is set and
is valid. The two latter functions will allow one to obtain a copy
of the respective receive and send keys. The return value of these
functions is the length of the valid key material. For security reasons,
one should take care to clear these copies when work is complete. The
max length of MPPE receive ands send keys are up to 32 bytes long, or
of MPPE_MAX_KEY_SIZE length.

The previous definitions of MPPE_MAX_KEY_LEN is the maximum length in
which the Linux kernel will accept for MPPE key lengths. Plugins would
access the MPPE keys directly via the:

  extern u_char mppe_send_key[MPPE_MAX_KEY_LEN]
  extern u_char mppe_recv_key[MPPE_MAX_KEY_LEN]

variables. The 2.5.0 release prohibits the direct access of these
variables by making them static and private in favor of using the new
API.

## $Id: PLUGINS,v 1.8 2008/06/15 07:02:18 paulus Exp $ ##