This guide will walk you through the basics of creating a new reference policy module. This will also serve as an introduction to the basic concepts and philosophy of refpolicy.

Modules are the principal organizing component in refpolicy. A module contains the policy for an application or related group of applications, private and shared resources, labeling information, and interfaces that allow other modules access to the module's resources. The majority of the global policy has been eliminated in refpolicy. Certain policy components, like users and object classes, are still global in refpolicy, but almost all TE policy is now contained within a module.

Let's create a new module called myapp. This is done by creating three files: myapp.te, myapp.fc, and myapp.if. The file myapp.te will contain all of the policy private to this module, including any types or attributes. The file myapp.fc file will contain the file context labeling statements for this module. Finally, the file myapp.if will contain the interfaces for this module (interfaces will be explained below).

First create myapp.te and add the following:

policy_module(myapp,1.0)

# Private type declarations
type myapp_t;
type myapp_exec_t;
type myapp_log_t;
type myapp_tmp_t;

domain_type(myapp_t)
domain_entry_file(myapp_t, myapp_exec_t)
logging_log_file(myapp_log_t)
files_tmp_file(myapp_tmp_t)

This creates all of the types needed for this module, including a type for the process, executables, log files, and temporary files. The first thing to notice is that there are no attributes applied to any of these types. In refpolicy all types and attributes can only be referred to in the module that declares them. This means that it is not possible, for example, to directly refer to the domain attribute. Instead, macros in other modules are used to declare that a type will be used for a certain purpose. These macros will likely use attributes (but not necessarily), but it allows the module that declared the attribute to strictly control how it can be used. In this example interfaces are used to transform the types into a domain, entry file, log file, and temporary file.

Let's expand this example further by allowing some access for these types. My application needs access between its own types and access to read random numbers. The access between private types is written exactly the same way current policy rules are written, i.e.:

allow myapp_t myapp_log_t:file append_file_perms;
allow myapp_t myapp_tmp_t:file manage_file_perms;

This allows myapp_t to write to its private types, but it needs to be able to create its temporary files in /tmp. This requires a call to the files module.

files_tmp_filetrans(myapp_t,myapp_tmp_t,file)

This call to the files module allows myapp_t to create myapp_tmp_t files in the /tmp directory.

The file contexts file lists files and the labels they should have. Create myapp.fc and add the following:

/usr/bin/myapp    --  gen_context(system_u:object_r:myapp_exec_t,s0)

The gen_context() macro has three parameters, the base SELinux label, the MLS level, and optionally the MCS category. When compiling a module, the macro will add the appropriate MLS/MCS part to the label when needed. In the above example, the file will have a level of s0 when compiled for a MLS policy, and no categories when compiled for MCS. If the level s5:c1 is needed for MLS, and category c0 is needed for MCS, the above line would become:

/usr/bin/myapp    --  gen_context(system_u:object_r:myapp_exec_t,s5:c1,c0)

Since the MCS policy has only one sensitivity (s0), this is automatically added by the gen_context() macro, and should not be added by the user. The raw category set must be specified, instead of the translated category set (e.g., "SystemLow-SystemHigh").

The interface file creates the macros that other modules will use to gain access to my resources. This allows the module that created the type or attribute to define appropriate uses. Additionally, it provides a single point for documentation. Create myapp.if and add the following:

## <summary>Myapp example policy</summary>
## <desc>
##  <p>
##      More descriptive text about myapp.  The desc
##      tag can also use p, ul, and ol
##      html tags for formatting.
##  </p>
##  <p>
##      This policy supports the following myapp features:
##      <ul>
##      <li>Feature A</li>
##      <li>Feature B</li>
##      <li>Feature C</li>
##      </ul>
##  </p>
## </desc>

########################################
## <summary>
##  Execute a domain transition to run myapp.
## </summary>
## <param name="domain">
##  <summary>
##  Domain allowed to transition.
##  </summary>
## </param>

interface(`myapp_domtrans',`
    gen_requires(`
        type myapp_t, myapp_exec_t;
    ')

    domtrans_pattern($1,myapp_exec_t,myapp_t)
')

########################################
## <summary>
##  Read myapp log files.
## </summary>
## <param name="domain">
##  <summary>
##  Domain allowed to read the log files.
##  </summary>
## </param>

interface(`myapp_read_log',`
    gen_requires(`
        type myapp_log_t;
    ')

    logging_search_logs($1)
    allow $1 myapp_log_t:file read_file_perms;
')

The first interface allows other domains to do a domain transition to myapp_t, by executing a program labeled myapp_exec_t.

The second interface allows other domains to read myapp's log files. Myapp's log files are in the /var/log directory, so the access to search the /var/log directory is also given by the interface. The gen_requires() macro is used to support loadable policy modules, and must explicitly list the type and attributes used by this interface. If object classes of a userland object manager are used, the class and the permissions used by the interface must also be listed.

Two methods of building modules are supported, headers and complete source. Current systems, that is Fedora Core 5 or newer, which support loadable policy modules should compile modules using headers. Using the complete source for building modules is only needed if loadable modules are not supported on the system or when doing other modifications to the base policy. Generally this is only suggested for experts.

When building a loadable policy module, the three module source files need not be in a specific directory. A development directory in the user's home directory would be sufficient. In this example, lets place it in the policy directory in the home directory. The example Makefile should be copied to this directory. It is usually located in the /usr/share/doc/PKGNAME directory, where PKGNAME is the name of the policy package that has the policy headers.

$ cp /usr/share/doc/refpolicy-20060307/Makefile.example ~/policy/Makefile

Alternatively, this can be copied from the Reference Policy source, from the doc directory. The Makefile is not required, but will simplify the process.

Now the policy directory should have the three module source files and Makefile. All that needs to be done is to run make, and the policy will be compiled.

$ make
Compiling targeted myapp module
/usr/bin/checkmodule:  loading policy configuration from tmp/myapp.tmp
/usr/bin/checkmodule:  policy configuration loaded
/usr/bin/checkmodule:  writing binary representation (version 5) to tmp/myapp.mod
Creating targeted myapp.pp policy package

If you do not have the example Makefile, you must tell make where to find the policy header's Makefile, by using the -f option. The Makefile for the base policy provided by the Linux distribution should be found in the /usr/share/selinux/NAME/include directory, where NAME is the name of the policy, for example, strict or targeted.

$ make -f /usr/share/selinux/targeted/include/Makefile
Compiling targeted myapp module
/usr/bin/checkmodule:  loading policy configuration from tmp/myapp.tmp
/usr/bin/checkmodule:  policy configuration loaded
/usr/bin/checkmodule:  writing binary representation (version 5) to tmp/myapp.mod
Creating targeted myapp.pp policy package

When this succeeds, there will be a myapp.pp policy package that can be inserted into the running policy. To load the module, you must be running as root, in a role allowed to run semodule. Then run semodule -i to insert the module into the running policy:

# semodule -i myapp.pp

The semodule command will only print messages if there is an error inserting the module. If it succeeds, semodule -l should list the myapp module, and the version:

# semodule -l
myapp   1.0