Whole document tree

4. The Linux-PAM configuration file

Linux-PAM is designed to provide the system administrator with a great deal of flexibility in configuring the privilege granting applications of their system. The local configuration of those aspects of system security controlled by Linux-PAM is contained in one of two places: either the single system file, /etc/pam.conf; or the /etc/pam.d/ directory. In this section we discuss the correct syntax of and generic options respected by entries to these files.

4.1 Configuration file syntax

The reader should note that the Linux-PAM specific tokens in this file are case insensitive. The module paths, however, are case sensitive since they indicate a file's name and reflect the case dependence of typical Linux file-systems. The case-sensitivity of the arguments to any given module is defined for each module in turn.

In addition to the lines described below, there are two special characters provided for the convenience of the system administrator: comments are preceded by a `#' and extend to the next end-of-line; also, module specification lines may be extended with a `\' escaped newline.

A general configuration line of the /etc/pam.conf file has the following form:

service-name   module-type   control-flag   module-path   arguments

service-name

The name of the service associated with this entry. Frequently the service name is the conventional name of the given application. For example, `ftpd', `rlogind' and `su', etc. .

There is a special service-name, reserved for defining a default authentication mechanism. It has the name `OTHER' and may be specified in either lower or upper case characters. Note, when there is a module specified for a named service, the `OTHER' entries are ignored.

module-type

One of (currently) four types of module. The four types are as follows:

• auth; this module type provides two aspects of authenticating the user. Firstly, it establishes that the user is who they claim to be, by instructing the application to prompt the user for a password or other means of identification. Secondly, the module can grant group membership (independently of the /etc/groups file discussed above) or other privileges through its credential granting properties.
• account; this module performs non-authentication based account management. It is typically used to restrict/permit access to a service based on the time of day, currently available system resources (maximum number of users) or perhaps the location of the applicant user---`root' login only on the console.
• session; primarily, this module is associated with doing things that need to be done for the user before/after they can be given service. Such things include the logging of information concerning the opening/closing of some data exchange with a user, mounting directories, etc. .
• password; this last module type is required for updating the authentication token associated with the user. Typically, there is one module for each `challenge/response' based authentication (auth) module-type.

control-flag

The control-flag is used to indicate how the PAM library will react to the success or failure of the module it is associated with. Since modules can be stacked (modules of the same type execute in series, one after another), the control-flags determine the relative importance of each module. The application is not made aware of the individual success or failure of modules listed in the `/etc/pam.conf' file. Instead, it receives a summary success or fail response from the Linux-PAM library. The order of execution of these modules is that of the entries in the /etc/pam.conf file; earlier entries are executed before later ones. As of Linux-PAM v0.60, this control-flag can be defined with one of two syntaxes.

The simpler (and historical) syntax for the control-flag is a single keyword defined to indicate the severity of concern associated with the success or failure of a specific module. There are four such keywords: required, requisite, sufficient and optional.

The Linux-PAM library interprets these keywords in the following manner:

• required; this indicates that the success of the module is required for the module-type facility to succeed. Failure of this module will not be apparent to the user until all of the remaining modules (of the same module-type) have been executed.
• requisite; like required, however, in the case that such a module returns a failure, control is directly returned to the application. The return value is that associated with the first required or requisite module to fail. Note, this flag can be used to protect against the possibility of a user getting the opportunity to enter a password over an unsafe medium. It is conceivable that such behavior might inform an attacker of valid accounts on a system. This possibility should be weighed against the not insignificant concerns of exposing a sensitive password in a hostile environment.
• sufficient; the success of this module is deemed `sufficient' to satisfy the Linux-PAM library that this module-type has succeeded in its purpose. In the event that no previous required module has failed, no more `stacked' modules of this type are invoked. (Note, in this case subsequent required modules are not invoked.). A failure of this module is not deemed as fatal to satisfying the application that this module-type has succeeded.
• optional; as its name suggests, this control-flag marks the module as not being critical to the success or failure of the user's application for service. In general, Linux-PAM ignores such a module when determining if the module stack will succeed or fail. However, in the absence of any definite successes or failures of previous or subsequent stacked modules this module will determine the nature of the response to the application. One example of this latter case, is when the other modules return something like PAM_IGNORE.

The more elaborate (newer) syntax is much more specific and gives the administrator a great deal of control over how the user is authenticated. This form of the control flag is delimeted with square brackets and consists of a series of value=action tokens:

    [value1=action1 value2=action2 ...]

Here, valueI is one of the following return values: success; open_err; symbol_err; service_err; system_err; buf_err; perm_denied; auth_err; cred_insufficient; authinfo_unavail; user_unknown; maxtries; new_authtok_reqd; acct_expired; session_err; cred_unavail; cred_expired; cred_err; no_module_data; conv_err; authtok_err; authtok_recover_err; authtok_lock_busy; authtok_disable_aging; try_again; ignore; abort; authtok_expired; module_unknown; bad_item; conv_again; incomplete; and default. The last of these (default) can be used to set the action for those return values that are not explicitly defined.

The actionI can be a positive integer or one of the following tokens: ignore; ok; done; bad; die; and reset. A positive integer, J, when specified as the action, can be used to indicate that the next J modules of the current type will be skipped. In this way, the administrator can develop a moderately sophisticated stack of modules with a number of different paths of execution. Which path is taken can be determined by the reactions of individual modules.

• ignore - when used with a stack of modules, the module's return status will not contribute to the return code the application obtains.
• bad - this action indicates that the return code should be thought of as indicative of the module failing. If this module is the first in the stack to fail, its status value will be used for that of the whole stack.
• die - equivalent to bad with the side effect of terminating the module stack and PAM immediately returning to the application.
• ok - this tells PAM that the administrator thinks this return code should contribute directly to the return code of the full stack of modules. In other words, if the former state of the stack would lead to a return of PAM_SUCCESS, the module's return code will override this value. Note, if the former state of the stack holds some value that is indicative of a modules failure, this 'ok' value will not be used to override that value.
• done - equivalent to ok with the side effect of terminating the module stack and PAM immediately returning to the application.
• reset - clear all memory of the state of the module stack and start again with the next stacked module.

Just to get a feel for the power of this new syntax, here is a taste of what you can do with it. With Linux-PAM-0.63, the notion of client plug-in agents was introduced. This is something that makes it possible for PAM to support machine-machine authentication using the transport protocol inherent to the client/server application. With the ``[ ... value=action ... ]'' control syntax, it is possible for an application to be configured to support binary prompts with compliant clients, but to gracefully fall over into an alternative authentication mode for older, legacy, applications. Flexible eh?

module-path

The path-name of the dynamically loadable object file; the pluggable module itself. If the first character of the module path is `/', it is assumed to be a complete path. If this is not the case, the given module path is appended to the default module path: /usr/lib/security (but see the notes above).

args

The args are a list of tokens that are passed to the module when it is invoked. Much like arguments to a typical Linux shell command. Generally, valid arguments are optional and are specific to any given module. Invalid arguments are ignored by a module, however, when encountering an invalid argument, the module is required to write an error to syslog(3). For a list of generic options see the next section.

Any line in (one of) the confiuration file(s), that is not formatted correctly, will generally tend (erring on the side of caution) to make the authentication process fail. A corresponding error is written to the system log files with a call to syslog(3).

4.2 Directory based configuration

More flexible than the single configuration file, as of version 0.56, it is possible to configure libpam via the contents of the /etc/pam.d/ directory. In this case the directory is filled with files each of which has a filename equal to a service-name (in lower-case): it is the personal configuration file for the named service.

Linux-PAM can be compiled in one of two modes. The preferred mode uses either /etc/pam.d/ or /etc/pam.conf configuration but not both. That is to say, if there is a /etc/pam.d/ directory then libpam only uses the files contained in this directory. However, in the absence of the /etc/pam.d/ directory the /etc/pam.conf file is used. The other mode (and the one currently supported by Red Hat 4.2 and higher) is to use both /etc/pam.d/ and /etc/pam.conf in sequence. In this mode, entries in /etc/pam.d/ override those of /etc/pam.conf.

The syntax of each file in /etc/pam.d/ is similar to that of the /etc/pam.conf file and is made up of lines of the following form:

module-type   control-flag   module-path   arguments

This method of configuration has a number of advantages over the single file approach. We list them here to assist the reader in deciding which scheme to adopt:

• A lower chance of misconfiguring an application. There is one less field to mis-type when editing the configuration files by hand.
• Easier to maintain. One application may be reconfigured without risk of interfering with other applications on the system.
• It is possible to symbolically link different services configuration files to a single file. This makes it easier to keep the system policy for access consistent across different applications. (It should be noted, to conserve space, it is equally possible to hard link a number of configuration files. However, care should be taken when administering this arrangement as editing a hard linked file is likely to break the link.)
• A potential for quicker configuration file parsing. Only the relevant entries are parsed when a service gets bound to its modules.
• It is possible to limit read access to individual Linux-PAM configuration files using the file protections of the filesystem.
• Package management becomes simpler. Every time a new application is installed, it can be accompanied by an /etc/pam.d/xxxxxx file.

4.3 Generic optional arguments

The following are optional arguments which are likely to be understood by any module. Arguments (including these) are in general optional.

debug

Use the syslog(3) call to log debugging information to the system log files.

no_warn

Instruct module to not give warning messages to the application.

use_first_pass

The module should not prompt the user for a password. Instead, it should obtain the previously typed password (from the preceding auth module), and use that. If that doesn't work, then the user will not be authenticated. (This option is intended for auth and password modules only).

try_first_pass

The module should attempt authentication with the previously typed password (from the preceding auth module). If that doesn't work, then the user is prompted for a password. (This option is intended for auth modules only).

use_mapped_pass

This argument is not currently supported by any of the modules in the Linux-PAM distribution because of possible consequences associated with U.S. encryption exporting restrictions. Within the U.S., module developers are, of course, free to implement it (as are developers in other countries). For compatibility reasons we describe its use as suggested in the DCE-RFC 86.0, see section bibliography for a pointer to this document.

The use_mapped_pass argument instructs the module to take the clear text authentication token entered by a previous module (that requests such a token) and use it to generate an encryption/decryption key with which to safely store/retrieve the authentication token required for this module. In this way the user can enter a single authentication token and be quietly authenticated by a number of stacked modules. Obviously a convenient feature that necessarily requires some reliably strong encryption to make it secure. This argument is intended for the auth and password module types only.

expose_account

In general the leakage of some information about user accounts is not a secure policy for modules to adopt. Sometimes information such as users names or home directories, or preferred shell, can be used to attack a user's account. In some circumstances, however, this sort of information is not deemed a threat: displaying a user's full name when asking them for a password in a secured environment could also be called being 'friendly'. The expose_account argument is a standard module argument to encourage a module to be less discrete about account information as it is deemed appropriate by the local administrator.

4.4 Example configuration file entries

In this section, we give some examples of entries that can be present in the Linux-PAM configuration file. As a first attempt at configuring your system you could do worse than to implement these.

Default policy

If a system is to be considered secure, it had better have a reasonably secure `OTHER' entry. The following is a paranoid setting (which is not a bad place to start!):

#
# default; deny access
#
OTHER   auth     required       /usr/lib/security/pam_deny.so
OTHER   account  required       /usr/lib/security/pam_deny.so
OTHER   password required       /usr/lib/security/pam_deny.so
OTHER   session  required       /usr/lib/security/pam_deny.so

The module pam_deny (documented in a later section) is not very sophisticated. For example, it logs no information when it is invoked so unless the users of a system contact the administrator when failing to execute a service application, the administrator may go for a long while in ignorance of the fact that his system is misconfigured.

The addition of the following line before those in the above example would provide a suitable warning to the administrator.

#
# default; wake up! This application is not configured
#
OTHER   auth     required       /usr/lib/security/pam_warn.so
OTHER   password required       /usr/lib/security/pam_warn.so

On a system that uses the /etc/pam.d/ configuration, the corresponding default setup would be achieved with the following file:

#
# default configuration: /etc/pam.d/other
#
auth     required       /usr/lib/security/pam_warn.so
auth     required       /usr/lib/security/pam_deny.so
account  required       /usr/lib/security/pam_deny.so
password required       /usr/lib/security/pam_warn.so
password required       /usr/lib/security/pam_deny.so
session  required       /usr/lib/security/pam_deny.so

On a less sensitive computer, one on which the system administrator wishes to remain ignorant of much of the power of Linux-PAM, the following selection of lines (in /etc/pam.conf) is likely to mimic the historically familiar Linux setup.

#
# default; standard UNIX access
#
OTHER   auth     required       /usr/lib/security/pam_unix_auth.so
OTHER   account  required       /usr/lib/security/pam_unix_acct.so
OTHER   password required       /usr/lib/security/pam_unix_passwd.so
OTHER   session  required       /usr/lib/security/pam_unix_session.so

To enable anonymous-ftp, the following lines might be used to replace the default (OTHER) ones. (*WARNING* as of 1996/12/28 this does not work correctly with any ftpd. Consequently, this description may be subject to change or the application will be fixed.)

#
# ftpd; add ftp-specifics. These lines enable anonymous ftp over
#       standard UNIX access (the listfile entry blocks access to
#       users listed in /etc/ftpusers)
#
ftpd    auth    sufficient  /usr/lib/security/pam_ftp.so
ftpd    auth    required    /usr/lib/security/pam_unix_auth.so use_first_pass
ftpd    auth    required    /usr/lib/security/pam_listfile.so \
                        onerr=succeed item=user sense=deny file=/etc/ftpusers