Scrutiny

The Scrutiny Framework provides an extensible plugin based architecture for building security auditing tools for Fuchsia. The primary goal is to have a common base for these tools to prevent code duplication. I encourage you to work through the Getting Started section to get a quick overview of the framework and how you can use it and write your own plugins!

Getting Started

To get started you can launch Scrutiny with the command fx scrutiny. If the tool isn't available on your path you can build it by simply adding: --with //src/security/scrutiny or --with //bundles:tools to your fx set command.

When you initially launch Scrutiny you will be greeted with a shell interface. This shell supports history, tab completions etc. The first command you will want to try out in the shell is:

help

This will list all of the currently available commands in Scrutiny. Each individual command can then be further dug into by using:

help command_name

With just these two instructions you should be able to get a pretty good idea of the different commands but to see some common workflows check out the examples directory which contains a selection of Scrutiny scripts (.sc).

Running Scripts

Scrutiny scripts (.sc) can be run with the fx scrutiny -s script_name command. This will output the content of your script to the screen as it runs through. To get started try out the following example:

fx scrutiny -s examples/zbi_inspect.sc

This script will extract the ZBI into your /tmp/ and allow you to see some of the inner workings of how the Zircon Boot Image is packaged.

Framework Architecture

Overview

This section of the guide provides a deep overview of the architecture behind the framework which should be useful to anyone looking to build a more complex plugin or who is looking to develop and improve the framework itself.

Data Model

The core of the framework revolves around the DataModel and is split into three categories:

  1. DataCollectors: Objects that collect data from the host or target and populate the DataModel. These transform raw data from the target or host device into the abstract data model that DataControllers consume to analyze the system.
  2. DataModel: This is an object relational mapping of Fuchsia. It provides a simple machine readable abstraction of the component topology along with other properties of the system state collected both statically and dynamically by the DataCollectors.
  3. DataControllers: Objects that analyze and collate data from the DataModel. This is where the complex analysis of the DataModel occurs completely independent of how the underlying data is collected. This provides flexability and compatability even if the underlying system changes.

Source: framework/src/model for the implementation of these three categories.

Plugin Management

Plugins take a set of DataCollectors and DataControllers and register them with the PluginManager. This is the primary interface you will work with when adding new features to the framework. Currently plugins must be built into the core binary.

A Scrutiny plugin has three core functions:

  1. Provide a mapping between a namespace and a DataController.
  2. Provide the set of DataCollectors required so that the DataModel is populated with all the information the DataControllers need to service queries.
  3. Provide a set of dependencies this plugin requires to operate correctly. This allows developers to build lightweight plugins that build ontop of the work of other plugins.

The abstract interface defined in framework/src/engine/plugin.rs allows the developer to focus on developing the plugin and can benefit from a shared architecture for integrating their plugin into both a REST service and a DataCollector worker pool.

REST Service

The REST service provides a way for DataVisualizers to access DataControllers over the network through REST JSON. This provides the greatest flexability as DataVisualizers can be written in any language or framework as long as they can communicate over the network to the Scrutiny service.

The REST service is populated automatically by the PluginManager. If a Plugin is registered and loaded and provides a PluginHook for a DataController mapped to the namespace “/foo/bar” the REST service will then call that DataController when the URL “http://localhost:8080/foo/bar” is accessed. It handles passing through the query string which is the HTTP body of the request.

This allows the developer to just work on the plugin and take advantage of the shared infrastructure for hooking into DataVisualizers.