Skip to content

A library to extract Code Property Graphs from C/C++, Java, Go, Python, Ruby and every other language through LLVM-IR.

License

Notifications You must be signed in to change notification settings

Fraunhofer-AISEC/cpg

This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.

Folders and files

NameName
Last commit message
Last commit date

Latest commit

1ce6e0c · Nov 29, 2024
Nov 29, 2024
Nov 29, 2024
Sep 30, 2024
Nov 21, 2024
Jul 15, 2024
Nov 29, 2024
Nov 29, 2024
Nov 29, 2024
Nov 29, 2024
Nov 6, 2024
Aug 29, 2024
Nov 18, 2024
Nov 29, 2024
Aug 29, 2024
Nov 27, 2024
Oct 17, 2024
Nov 18, 2024
Nov 29, 2024
Aug 13, 2020
Jul 18, 2024
Nov 8, 2022
Sep 12, 2023
Dec 5, 2019
Nov 29, 2024
Mar 22, 2024
Nov 29, 2024
Sep 30, 2024
Nov 29, 2024
Nov 29, 2024
Jul 12, 2024
Jul 12, 2024
Jun 4, 2024
Nov 29, 2024
Sep 12, 2023

Repository files navigation

Code Property Graph

Actions Status codecov

A simple library to extract a code property graph out of source code. It has support for multiple passes that can extend the analysis after the graph is constructed. It currently supports C/C++ (C17), Java (Java 13) and has experimental support for Golang, Python and TypeScript. Furthermore, it has support for the LLVM IR and thus, theoretically support for all languages that compile using LLVM.

What is this?

A code property graph (CPG) is a representation of source code in form of a labelled directed multi-graph. Think of it as directed a graph where each node and edge is assigned a (possibly empty) set of key-value pairs (properties). This representation is supported by a range of graph databases such as Neptune, Cosmos, Neo4j, Titan, and Apache Tinkergraph and can be used to store source code of a program in a searchable data structure. Thus, the code property graph allows to use existing graph query languages such as Cypher, NQL, SQL, or Gremlin in order to either manually navigate through interesting parts of the source code or to automatically find "interesting" patterns.

This library uses Eclipse CDT for parsing C/C++ source code JavaParser for parsing Java. In contrast to compiler AST generators, both are "forgiving" parsers that can cope with incomplete or even semantically incorrect source code. That makes it possible to analyze source code even without being able to compile it (due to missing dependencies or minor syntax errors). Furthermore, it uses LLVM through the javacpp project to parse LLVM IR. Note that the LLVM IR parser is not forgiving, i.e., the LLVM IR code needs to be at least considered valid by LLVM. The necessary native libraries are shipped by the javacpp project for most platforms.

Specifications

In order to improve some formal aspects of our library, we created several specifications of our core concepts. Currently, the following specifications exist:

We aim to provide more specifications over time.

Usage

To build the project from source, you have to generate a gradle.properties file locally. This file also enables and disables the supported programming languages. We provide a sample file here - simply copy it to gradle.properties in the directory of the cpg-project. Instead of manually generating or editing the gradle.properties file, you can also use the configure_frontends.sh script, which edits the properties setting the supported programming languages for you.

For Visualization Purposes

In order to get familiar with the graph itself, you can use the subproject cpg-neo4j. It uses this library to generate the CPG for a set of user-provided code files. The graph is then persisted to a Neo4j graph database. The advantage this has for the user, is that Neo4j's visualization software Neo4j Browser can be used to graphically look at the CPG nodes and edges, instead of their Java representations.

As Library

The most recent version is being published to Maven central and can be used as a simple dependency, either using Maven or Gradle.

dependencies {
    val cpgVersion = "8.0.0"

    // if you want to include all published cpg modules
    implementation("de.fraunhofer.aisec", "cpg", cpgVersion)

    // if you only want to use some of the cpg modules
    // use the 'cpg-core' module
    // and then add the needed extra modules, such as Go and Python
    implementation("de.fraunhofer.aisec", "cpg-core", cpgVersion)
    implementation("de.fraunhofer.aisec", "cpg-language-go", cpgVersion)
    implementation("de.fraunhofer.aisec", "cpg-language-python", cpgVersion)
}

There are some extra steps necessary for the cpg-language-cxx module. Since Eclipse CDT is not published on maven central, it is necessary to add a repository with a custom layout to find the released CDT files. For example, using Gradle's Kotlin syntax:

repositories {
    // This is only needed for the C++ language frontend
    ivy {
        setUrl("https://download.eclipse.org/tools/cdt/releases/")
        metadataSources {
            artifact()
        }

        patternLayout {
            artifact("[organisation].[module]_[revision].[ext]")
        }
    }
}

Beware, that the cpg module includes all optional features and might potentially be HUGE (especially because of the LLVM support). If you do not need LLVM, we suggest just using the cpg-core module with the needed extra modules like cpg-language-go. In the future we are working on extracting more optional modules into separate modules.

Development Builds

A published artifact of every commit can be requested through JitPack. This is especially useful, if your external project makes use of a specific feature that is not yet merged in yet or not published as a version yet. Please follow the instructions on the JitPack page. Please be aware, that similar to release builds, the CDT repository needs to be added as well (see above).

On Command Line

The library can be used on the command line using the cpg-console subproject. Please refer to the README.md of the cpg-console as well as our small tutorial for further details.

Configuration

The behavior of the library can be configured in several ways. Most of this is done through the TranslationConfiguration and the InferenceConfiguration.

TranslationConfiguration

The TranslationConfiguration configures various aspects of the translation. E.g., it determines which languages/language frontends and passes will be used, which information should be inferred, which files will be included, among others. The configuration is set through a builder pattern.

InferenceConfiguration

The class InferenceConfiguration can be used to affect the behavior or the passes if they identify missing nodes. Currently, there are three flags which can be enabled:

  • guessCastExpression enables guessing if a CPP expression is a cast or a call expression if it is not clear.
  • inferRecords enables the inference of missing record declarations (i.e., classes and structs)
  • inferDfgForUnresolvedSymbols adds DFG edges to method calls represent all potential data flows if the called function is not present in the source code under analysis.

Only inferDfgForUnresolvedSymbols is turned on by default.

The configuration can be made through a builder pattern and is set in the TranslationConfiguration as follows:

val inferenceConfig = InferenceConfiguration
    .builder()
    .guessCastExpression(true)
    .inferRecords(true)
    .inferDfgForUnresolvedSymbols(true)
    .build()

val translationConfig = TranslationConfiguration
    .builder()
    .inferenceConfiguration(inferenceConfig)
    .build()

Development

This section describes languages, how well they are supported, and how to use and develop them yourself.

Language Support

Languages are maintained to different degrees, and are noted in the table below with:

  • maintained: if they are mostly feature complete and bugs have priority of being fixed.
  • incubating: if the language is currently being worked on to reach a state of feature completeness.
  • experimental: if a first working prototype was implemented, e.g., to support research topics, and its future development is unclear.
  • discontinued: if the language is no longer actively developed or maintained but is kept for everyone to fork and adapt.

The current state of languages is:

Language Module Branch State
Java (Source) cpg-language-java main maintained
C++ cpg-language-cxx main maintained
Python cpg-language-python main maintained
Go cpg-language-go main maintained
INI cpg-language-ini main maintained
JVM (Bytecode) cpg-language-jvm main incubating
LLVM cpg-language-llvm main incubating
TypeScript/JavaScript cpg-language-typescript main experimental
Ruby cpg-language-ruby main experimental
{OpenQASM,Python-Qiskit} cpg-language-{openqasm,python-qiskit} quantum-cpg experimental

Languages and Configuration

cpg-core contains the graph nodes, language-independent passes that add semantics to the cpg-AST. Languages are developed in separate gradle submodules. To include the desired language submodules, simply toggle them on in your local gradle.properties file by setting the properties to true, e.g., (enableGoFrontend=true). We provide a sample file with all languages switched on here. Instead of manually editing the gradle.properties file, you can also use the configure_frontends.sh script, which edits the properties for you. Some languages need additional installation of software to run and will be listed below.

Golang

In the case of Golang, additional native code, libgoast, is used to access the Go ast packages. Gradle should automatically download the latest version of this library during the build process. This currently only works for Linux and macOS.

Python

You need to install jep. This can either be system-wide or in a virtual environment. Your jep version has to match the version used by the CPG (see version catalog).

Currently, only Python 3.{9,10,11,12,13} is supported.

System Wide

Follow the instructions at https://github.com/ninia/jep/wiki/Getting-Started#installing-jep.

Virtual Env
  • python3 -m venv ~/.virtualenvs/cpg
  • source ~/.virtualenvs/cpg/bin/activate
  • pip3 install jep

Through the JepSingleton, the CPG library will look for well known paths on Linux and OS X. JepSingleton will prefer a virtualenv with the name cpg, this can be adjusted with the environment variable CPG_PYTHON_VIRTUALENV.

TypeScript

For parsing TypeScript, the necessary NodeJS-based code can be found in the src/main/nodejs directory of the cpg-language-typescript submodule. Gradle should build the script automatically, provided NodeJS (>=16) is installed. The bundles script will be placed inside the jar's resources and should work out of the box.

Code Style

We use Google Java Style as a formatting. Please install the appropriate plugin for your IDE, such as the google-java-format IntelliJ plugin or google-java-format Eclipse plugin.

Integration into IntelliJ

Straightforward, however three things are recommended

  • Enable gradle "auto-import"
  • Enable google-java-format
  • Hook gradle spotlessApply into "before build" (might be obsolete with IDEA 2019.1)

Git Hooks

You can use the hook in style/pre-commit to check for formatting errors:

cp style/pre-commit .git/hooks

Contributors

The following authors have contributed to this project (in alphabetical order):

Contributing

Before accepting external contributions, you need to sign our CLA. Our CLA assistent will check, whether you already signed the CLA when you open your first pull request.

Further reading

You can find a complete list of papers here

A quick write-up of our CPG has been published on arXiv:

[1] Konrad Weiss, Christian Banse. A Language-Independent Analysis Platform for Source Code. https://arxiv.org/abs/2203.08424

A preliminary version of this cpg has been used to analyze ARM binaries of iOS apps:

[2] Julian Schütte, Dennis Titze. liOS: Lifting iOS Apps for Fun and Profit. Proceedings of the ESORICS International Workshop on Secure Internet of Things (SIoT), Luxembourg, 2019. https://arxiv.org/abs/2003.12901

An initial publication on the concept of using code property graphs for static analysis:

[3] Yamaguchi et al. - Modeling and Discovering Vulnerabilities with Code Property Graphs. https://www.sec.cs.tu-bs.de/pubs/2014-ieeesp.pdf

[4] is an unrelated, yet similar project by the authors of the above publication, that is used by the open source software Joern [5] for analysing C/C++ code. While [4] is a specification and implementation of the data structure, this project here includes various Language frontends (currently C/C++ and Java, Python to com) and allows creating custom graphs by configuring Passes which extend the graph as necessary for a specific analysis:

[4] https://github.com/ShiftLeftSecurity/codepropertygraph

[5] https://github.com/ShiftLeftSecurity/joern/

Additional extensions of the CPG to support further use-cases:

[6] Christian Banse, Immanuel Kunz, Angelika Schneider and Konrad Weiss. Cloud Property Graph: Connecting Cloud Security Assessments with Static Code Analysis. IEEE CLOUD 2021. https://doi.org/10.1109/CLOUD53861.2021.00014

[7] Alexander Küchler, Christian Banse. Representing LLVM-IR in a Code Property Graph. 25th Information Security Conference (ISC). Bali, Indonesia. 2022

[8] Maximilian Kaul, Alexander Küchler, Christian Banse. A Uniform Representation of Classical and Quantum Source Code for Static Code Analysis. IEEE International Conference on Quantum Computing and Engineering (QCE). Bellevue, WA, USA. 2023