For the uninitiated, we begin by considering an example. We take an application that grants some service to users; login is one such program. Login does two things, it first establishes that the requesting user is whom they claim to be and second provides them with the requested service: in the case of login the service is a command shell (bash, tcsh, zsh, etc.) running with the identity of the user.
Traditionally, the former step is achieved by the login application prompting the user for a password and then verifying that it agrees with that located on the system; hence verifying that as far as the system is concerned the user is who they claim to be. This is the task that is delegated to Linux-PAM.
From the perspective of the application programmer (in this case the person that wrote the login application), Linux-PAM takes care of this authentication task -- verifying the identity of the user.
The flexibility of Linux-PAM is that you, the system
administrator, have the freedom to stipulate which authentication
scheme is to be used. You have the freedom to set the scheme for
any/all PAM-aware applications on your Linux system. That is, you can
authenticate from anything as naive as simple trust
(pam_permit) to something as paranoid as a combination of a
retinal scan, a voice print and a one-time password!
To illustrate the flexibility you face, consider the following situation: a system administrator (parent) wishes to improve the mathematical ability of her users (children). She can configure their favorite ``Shoot 'em up game'' (PAM-aware of course) to authenticate them with a request for the product of a couple of random numbers less than 12. It is clear that if the game is any good they will soon learn their multiplication tables. As they mature, the authentication can be upgraded to include (long) division!
Linux-PAM deals with four separate types of (management) task. These are: authentication management; account management; session management; and password management. The association of the preferred management scheme with the behavior of an application is made with entries in the relevant Linux-PAM configuration file. The management functions are performed by modules specified in the configuration file. The syntax for this file is discussed in the section below.
Here is a figure that describes the overall organization of Linux-PAM.
+----------------+
| application: X |
+----------------+ / +----------+ +================+
| authentication-[---->--\--] Linux- |--<--| PAM config file|
| + [----<--/--] PAM | |================|
|[conversation()][--+ \ | | | X auth .. a.so |
+----------------+ | / +-n--n-----+ | X auth .. b.so |
| | | __| | | _____/
| service user | A | | |____,-----'
| | | V A
+----------------+ +------|-----|---------+ -----+------+
+---u-----u----+ | | |
| auth.... |--[ a ]--[ b ]--[ c ]
+--------------+
| acct.... |--[ b ]--[ d ]
+--------------+
| password |--[ b ]--[ c ]
+--------------+
| session |--[ e ]--[ c ]
+--------------+
The following text was contributed by Seth Chaiklin:
To this point, we have described how PAM should work in an
ideal world, in which all applications are coded properly.
However, at the present time (October 1998), this is far
from the case. Therefore, here are some practical considerations
in trying to use PAM in your system.
Why bother, is it really worth all the trouble?
If you running Linux as a single user system, or in an
environment where all the users are trusted, then there
is no real advantage for using PAM.
Ed: there is actually an advantage since you can dummy down the authentication to the point where you don't have any... Almost like Win95.
In a networked environment, it is clear that you need to think a little more about how users etc., are authenticated:]
If you are running Linux as a server, where several different
services are being provided (e.g., WWW with areas restricted by
password control, PPP), then there can be some real and interesting
value for PAM. In particular, through the use of modules, PAM can
enable a program to search through several different password
databases, even if that program is not explicitly coded for
that particular database. Here are some examples of the possibilities
that this enables.
o Apache has a module that provides PAM services. Now
authentication
to use particular directories can be conducted by PAM, which
means that the range of modules that are available to PAM can
be used, including RADIUS, NIS, NCP (which means that Novell
password databases can be used).
o pppd has a PAMified version (available from RedHat) Now it is
possible to use a series of databases to authenticate ppp users.
In addition to the normal Linux-based password databases (such
as /etc/passwd and /etc/shadow), you can use PAM modules to
authenticate against Novell password databases or NT-based
password databases.
o The preceding two examples can be combined. Imagaine that the
persons in your office/department are already registered with a
username and password in a Novell or NT LAN. If you wanted to
use this database on your Linux server (for PPP access, for
web access, or even for normal shell access), you can use PAM
to authenticate against this existing database, rather than
maintain a separate database on both Linux and the LAN server.
Can I use PAM for any program that requires authentication?
Yes and no. Yes, if you have access to the source code, and can
add the appropriate PAM functions. No, if you do not have access
to the source code, and the binary does not have the PAM functions
included.
In other words, if a program is going to use PAM, then it has to
have PAM functions explicitly coded into the program. If they
are not, then it is not possible to use PAM.
How can I tell whether a program has PAM coded into it or not?
A quick-and-dirty (but not always reliable) method is to ldd
<programname>
If libpam and libpam_misc are not among the libraries that the program
uses, then it is not going to work with PAM. However, it is possible
that the libraries are included, but there are still problems, because
the PAM coding in the program does not work as it should. So a
more reliable method is to make the follow tests.
In the /etc/pam.d directory, one needs to make a configuration file
for the program that one wants to run. The exact name of the
configuration
file is hard-coded into the program. Usually, it is the same name as
the
program, but not always. For sake of illustration, let's assume that
the program is named "pamprog" and the name of the configuration file
is /etc/pam.d/pamprog.
In the /etc/pam.d/pamprog put the following two lines:
auth required pam_permit.so
auth required pam_warn.so
Now try to use pamprog. The first line in the configuration file
says that all users are permitted. The second line will write a
warning to your syslog file (or whether you syslog is writing
messages). If this test succeeds, then you know that you have
a program that can understand pam, and you can start the more
interesting work of deciding how to stack modules in your
/etc/pam.d/pamprog file.