-
Notifications
You must be signed in to change notification settings - Fork 762
Authorization
This howto provides information about opengrok authorization with authorization framework. You can use this module along with project groupings which allows you to create various project structures.
If plugin directory does not exist or if it contains no plugins the framework just allows every request.
You can configure:
-
Plugin directory
The configuration has been moved from
web.xml
to the usual OpenGrok configuration XML file. The way you provide new plugin directory is by creating a new XML configuration "read-only configuration" file which gets merged with the configuration made by the indexer.There in the file provide a new variable named
pluginDirectory
with aString
parameter with absolute path to the directory with plugins.By default the directory points to
OPENGROK_DATA_ROOT
/plugins. TheOPENGROK_DATA_ROOT
is thedataRoot
directory set in the mainconfiguration.xml
. If no plugin is available/the directory does not exist then the framework allows every request.Opengrok authorization expects plugin classes in that directory, it loads those classes in memory as plugins and then perform the authorization for each request.
-
Watchdog service
Similar for the watchdog service there is a dedicated option in the configuration which you can set as described above. The name of it is
authorizationWatchdogEnabled
and the value is aboolean
type.This service watches for changes (even recursive) in the plugin directory and if such change was done (modifing, deleting); it reloads the plugins. All old plugins in memory are discarded and only the plugins contained in plugin directory are loaded. This is done in real time
NOTE: Modification of every file in plugin directory is propagated as an event. This means that modifying bunch of files in the plugin directory leads to several reload events
By default the value is
false
. -
Authorization stack
Discussed under advanced configuration.
-
Provide a plugin
You can provide your own plugin by putting the compiled plugin class (.class or in .jar) into plugin directory.
NOTE: The important experience came out when using .jar files and is discussed in troubleshooting section.
Example of a TruePlugin.java is included in plugins directory in the root of the git repository.
Each plugin must implement the IAuthorizationPlugin
interface which contains four methods:
-
void load ( void );
This method is called whenever the framework loads the plugin into memory
- watchdog service enabled and change was made (forced reload)
- deploy of the webapp
- configuration change (change of the plugin directory, ...)
-
void unload ( void );
This method is called whenever the framework destroys the plugin
- watchdog service enabled and change was made (forced reload) (the order is first unload the old instance then load the new one)
- undeploy of the webapp
-
boolean isAllowed(HttpServletRequest request, Project project);
This method gives a decision if the project should be allowed for the current request.
-
boolean isAllowed(HttpServletRequest request, Group group);
This analogically gives the same decision for the group.
All those methods are also described in the code.
Each is expected to implement some sort of a cache for the decisions when the underlying read operations for the user rights is expensive.
HttpServletRequest
object is the current request with all of its features
like: session, attributes, getUser()
, getPrincipal()
If you decide that your plugin uses some sort of sessions then you might want to reload the session as well when the authorization framework reloads your plugin. The framework tracks a number of reloads which is available for you as RuntimeEnvironment.getInstance().getPluginVersion() and you can check this number and invalidate your session if your version (tracked perhaps in your session) is different than this.
Custom classloader restricts the plugin to load only this classes from org.opengrok.indexer
package:
private final static String[] classWhitelist = new String[]{
"org.opengrok.indexer.configuration.Group",
"org.opengrok.indexer.configuration.Project",
"org.opengrok.indexer.configuration.RuntimeEnvironment",
"org.opengrok.indexer.authorization.IAuthorizationPlugin",
"org.opengrok.indexer.util.*",
"org.opengrok.indexer.logger.*",
};
And explicitly forbids these packages to be extended.
private final static String[] packageBlacklist = new String[]{
"java",
"javax",
"org.w3c",
"org.xml",
"org.omg",
"sun"
};
Also JVM can forbid you to extend some packages which are not meant to be extended.
The plugin class must be compiled to the .class file (and the it can be packaged into .jar file). The frameworks supports both .class and .jar files. For compiling you have to provide opengrok.jar
and the servlet api (HttpServletRequest
) in the classpath
Example (for the TruePlugin
which is included in the repository):
$ javac -classpath dist/opengrok.jar -d . plugins/TruePlugin.java
Then you can just drop the compiled .class
file into plugin directory and deploy the webapp.
If the plugin is a part of a package. Then you have to copy the full directory path which is made
by the compiler relatively to the source file.
If anything goes wrong, you should find information in the logs.
The framework consists of three parts
-
PluginFramework
This is the plugin container
- performs the authorization decision
- cache the decisions so that for each request the particular
plugin's
isAllowed
for particular project/group is called only once
-
ProjectHelper
UI facade for filtered (authorized) projects/groups/repositories which should be ONLY used for displaying filtered information anywhere
- provides methods how to get filtered projects/groups/repositories
- cache the filtered results for each request so that it does not have to call the framework unnecessarily
-
AuthorizationFilter
Standard
javax.servlet.Filter
which is used for all urls to decide if the current request has access to the url- restricts user to go to forbidden xref/history/download etc. with 403 HTTP error
- reacts only when the url contains information about project (so it can decide per project)
Every 403 error is logged. But information about the user is missing because the decision is made by plugins and the filter does not know how to interpret the request to get the user from it.
The new 1.0 release contain a new feature about how to configure the plugin invocation in more details.
The plugins form a linear structure of a list (called stack). The order of invocation of the plugins methods isAllowed
is determined by their position in the stack. Moreover you can configure a different flag for each of your plugins which is respected when performing the authorization.
There are three flags:
-
REQUIRED Failure of such a plugin will ultimately lead to the authorization framework returning failure but only after the remaining plugins have been invoked.
-
REQUISITE Like required, however, in the case that such a plugin returns a failure, control is directly returned to the application. The return value is that associated with the first required or requisite plugin to fail.
-
SUFFICIENT If such a plugin succeeds and no prior required plugin has failed the authorization framework returns success to the application immediately without calling any further plugins in the stack. A failure of a sufficient plugin is ignored and processing of the plugin list continues unaffected.
These are inspired by the PAM Authorization framework and the definition is taken directly from the man pages of the PAM configuration. However OpenGrok does not implement all PAM flags.
You can use the authorization framework without providing such an advanced configuration because all loaded plugins which do not occur in this advanced configuration are appended to the list with REQUIRED flag. However, as of the nature of the class discovery this means that the order of invocation of these plugins is rather random.
This is an example entry for the "read-only configuration". This defines three plugins, each of them has a different role and so affect the stack processing in different way. Use the class name to specify the targeted plugin and one of the flags as the authorization role. This stack is then for every request processed from the top to the bottom (as it is a list) evaluating the flags with the particular plugin decisions.
<void property="pluginStack">
<void method="add">
<object class="org.opengrok.indexer.authorization.AuthorizationPlugin">
<void property="flag">
<string>REQUISITE</string>
</void>
<void property="name">
<string>some.my.package.RequisitePlugin</string>
</void>
</object>
</void>
<void method="add">
<object class="org.opengrok.indexer.authorization.AuthorizationPlugin">
<void property="flag">
<string>SUFFICIENT</string>
</void>
<void property="name">
<string>Plugin</string>
</void>
</object>
</void>
<void method="add">
<object class="org.opengrok.indexer.authorization.AuthorizationPlugin">
<void property="flag">
<string>REQUIRED</string>
</void>
<void property="name">
<string>ExamplePlugin</string>
</void>
</object>
</void>
</void>
A typical use of this configuration would be:
RequisitePlugin
- Determine a user identity in the requisite plugin and store it in the request
- Return true or false if such action was successful
Plugin
- Use the user identity determined in the requisite plugin which is stored in the request
- Perform an authorization check which can lead to immediately return from the stack with a success value
- If the check is negative, the stack continues to the third plugin
ExamplePlugin
- Use the user identity determined in the requisite plugin which is stored in the request
- Filter the rest of which was not enabled by the previous
Plugin
implementation
This is an example using HTTP Basic authorization.
This example has several presumptions:
- you have a clean working OpenGrok instance on your machine (no projects, no groups)
- you are able to use
Groups
subcommand (available when you clone the github repository under tools) - you have the permission to modify Tomcat files
- you have set the plugin directory or you know where it is (described above)
HTTP Basic is supported by Tomcat and we can configure a simple example including users and roles. How we can set up the Tomcat is described in this tutorial using users and roles defined by us.
We have to modify a Tomcat file to provide information about users and roles in the system. This file is placed in $CATALINA_BASE/conf/tomcat-users.xml
.
For purpose of this example add these lines to the file inside the element <tomcat-users>
.
<role rolename="users"/>
<role rolename="admins"/>
<role rolename="plugins"/>
<role rolename="ghost"/>
<user username="007" password="123456" roles="users"/>
<user username="008" password="123456" roles="plugins"/>
<user username="009" password="123456" roles="users,admins"/>
<user username="00A" password="123456" roles="admins"/>
<user username="00B" password="123456" roles="admins,plugins"/>
<user username="00F" password="123456" roles="ghost"/>
With these lines we tell Tomcat to set up 4 roles and 6 users with their roles assigned. You can modify the usernames and passwords to fit your needs but we will use those in this example.
Now we have to tell the application that it should use HTTP Basic authentication to protect its sources. We can do this by modifying a web.xml
file which is usually placed in the WEB-INF
directory in your application. Following lines are necessary to get it work.
<security-constraint>
<web-resource-collection>
<web-resource-name>API endpoints are checked separately by the web app</web-resource-name>
<url-pattern>/api/*</url-pattern>
</web-resource-collection>
</security-constraint>
<security-constraint>
<web-resource-collection>
<web-resource-name>In general everything needs to be authenticated</web-resource-name>
<url-pattern>/*</url-pattern> <!-- protect the whole application -->
<url-pattern>/api/v1/search</url-pattern> <!-- protect search endpoint whitelisted above -->
<url-pattern>/api/v1/suggest/*</url-pattern> <!-- protect suggest endpoint whitelisted above -->
</web-resource-collection>
<auth-constraint>
<role-name>plugins</role-name> <!-- these are the roles from tomcat-users.xml -->
<role-name>users</role-name> <!-- these are the roles from tomcat-users.xml -->
<role-name>admins</role-name> <!-- these are the roles from tomcat-users.xml -->
</auth-constraint>
<user-data-constraint>
<!-- transport-guarantee can be CONFIDENTIAL, INTEGRAL, or NONE -->
<transport-guarantee>NONE</transport-guarantee>
</user-data-constraint>
</security-constraint>
<security-role>
<role-name>plugins</role-name>
<role-name>users</role-name>
<role-name>admins</role-name>
</security-role>
<login-config>
<auth-method>BASIC</auth-method>
</login-config>
We would strongly recommend you to turn on the watchdog service which is suitable for developing plugins, setting the parameter value to true
via the "read-only configuration".
This forces the application to reload all plugins in the plugin directory when a modification of any file inside it occurs.
We need to create a couple of test repositories to show the authorization features.
$ cd "$OPENGROK_SRC_ROOT" # navigate to the source folder
$ for name in `seq 1 11`; do mkdir "test-project-$name"; cd "test-project-$name"; echo "Give it a try! Hello from $i" > README.md; git init; git config user.email "x"; git config user.name "y"; git add .; git commit -m "init commit"; cd ..; done;
This should create 10 repositories named "test-project-$number" in the source directory.
See https://github.com/OpenGrok/OpenGrok/wiki/Project-groups
The group names correspond to the roles defined in tomcat-users.xml
earlier.
The final group structure should look like this now:
$ groups.py -a opengrok.jar -- -l -i readonly_configuration.xml
admins ~ "test-project-1|test-project-2|test-project-3|test-project-4"
users ~ "test-project-5|test-project-6|test-project-7|test-project-8"
plugins ~ "test-project-9|test-project-10"
Now run the index as usual. Do not forget to use the -R
option with group configuration as that will pass the groups into the main configuration. This is the only option how to preserve groups on reindex.
Now comes the main part - the plugin itself.
It consists of three parts:
- Permission policy
- Group discovery
- Authorization check
You can use whatever policy you want using even external tools in java or you OS. In this example we will use static map defining which user or group has access to which projects.
private static final Map<String, Set<String>> userProjects = new TreeMap<>();
private static final Map<String, Set<String>> userGroups = new TreeMap<>();
static {
// all have access to "test-project-11" and some to other "test-project-5" or "test-project-8"
userProjects.put("007", new TreeSet<>(Arrays.asList(new String[]{"test-project-11", "test-project-5"})));
userProjects.put("008", new TreeSet<>(Arrays.asList(new String[]{"test-project-11", "test-project-8"})));
userProjects.put("009", new TreeSet<>(Arrays.asList(new String[]{"test-project-11"})));
userProjects.put("00A", new TreeSet<>(Arrays.asList(new String[]{"test-project-11"})));
userProjects.put("00B", new TreeSet<>(Arrays.asList(new String[]{"test-project-11"})));
}
static {
userGroups.put("007", new TreeSet<>(Arrays.asList(new String[]{})));
userGroups.put("008", new TreeSet<>(Arrays.asList(new String[]{})));
userGroups.put("009", new TreeSet<>(Arrays.asList(new String[]{})));
userGroups.put("00A", new TreeSet<>(Arrays.asList(new String[]{})));
userGroups.put("00B", new TreeSet<>(Arrays.asList(new String[]{})));
}
The plugin framework which works above our plugin has no idea how specific our plugins wants to be when it does the decisions so there is no way of automate discovery of groups/subgroups/projects in groups. All has to be done in our plugin, preferably only once the plugin is loaded/first used for the particular user (not included in this example).
The very important implication is that allowing a group does not allow its subgroups neither projects. Consider this as a feature since it gives you more freedom for allowing or disallowing particular groups and projects.
This is the most important part - if you have found a group for the user then add all of its projects, repositories, subgroups and their underlying objects.
if ((g = Group.getByName(group)) != null) {
// group discovery
for (Project p : g.getRepositories()) {
userProjects.get(request.getUserPrincipal().getName()).add(p.getDescription());
}
for (Project p : g.getProjects()) {
userProjects.get(request.getUserPrincipal().getName()).add(p.getDescription());
}
for (Group grp : g.getDescendants()) {
for (Project p : grp.getRepositories()) {
userProjects.get(request.getUserPrincipal().getName()).add(p.getDescription());
}
for (Project p : grp.getProjects()) {
userProjects.get(request.getUserPrincipal().getName()).add(p.getDescription());
}
descendants.add(grp.getName());
}
while (g != null) {
descendants.add(g.getName());
g = g.getParent();
}
}
Eventually, the userProjects
and userGroups
are maps of type User-Projects/Groups where we can quickly decide whether the user has the permission for the given entity
by just looking into the map and the corresponding set.
The final authorization check is just simple - check if the map contains such Project/Group for the certain user.
// for projects
return userProjects.get(request.getUserPrincipal().getName()).contains(project.getDescription());
// or for groups
return userGroups.get(request.getUserPrincipal().getName()).contains(group.getName());
Now you can compile the plugin as described above in this wiki page and place it into the plugin directory (default $OPENGROK_DATA_ROOT/plugins
).
If you enabled watchdog service then you are ready to see the results, otherwise you need to restart the application.
When you enter the application, the page immediately fires an login form where you can enter the credentials (as written in tomcat users). Depending on what you have entered you should see filtered results on the main page and even when you search for anything you should not be able to access other projects.
You can try other accounts by logging out by forgetting the session (ctrl+shift+delete
+ forget current session/this can vary in different browsers).
HttpBasicAuthorizationPlugin.java
When using IDE (NetBeans) to build your plugin; you can face a problem when the framework does not load your .jar file with ClassFormatError
resulting to ClassNotFoundException
.
Possible solutions are:
- disable debugging symbols (project/properities/build/compile/uncheck generate debugging info)
- compile the .java files with
javac
manually and then package it into .jar manually (command above) - compile the .java files with
javac
manually and use this directory structure (without packaging)
This should solve the problem. If it does not then try to change the code.