LLFI is an LLVM based fault injection tool, that injects faults into the LLVM IR of the application source code. The faults can be injected into specific program points, and the effect can be easily tracked back to the source code. Please refer to the paper below. NOTE: If you publish a paper using LLFI, please add it to PaperLLFI.bib
C C++ Python Java JavaScript CSS Other
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.



LLFI is an LLVM based fault injection tool, that injects faults into the LLVM IR of the application source code. The faults can be injected into specific program points, and the effect can be easily tracked back to the source code. LLFI is typically used to map fault characteristics back to source code, and hence understand source level or program characteristics for various kinds of fault outcomes. Detailed documentation about LLFI can be found at: https://github.com/DependableSystemsLab/LLFI/wiki

Please join the following Google Groups for asking questions about LLFI that are not answered in the documentation: llfi-development@googlegroups.com


This is the recommended method for building the LLFI. If you wish to build the LLFI via the auto-installer, you do not need to clone the LLFI git repository. Simply download the installer script by itself, and it will fetch the latest version of the git repository for you. The LLFI auto-installer takes the form of a single python script (installer/installLLFI.py). To run the script, simply copy it into the directory where you would like to build the LLFI and, from the command line, run python3 InstallLLFI.py.


  1. 64 Bit Machine
  2. 64 Bit Linux or OS X
  3. Cmake (mininum v2.8)
  4. Python 3 and above
  5. tcsh (for GUI)
  6. GraphViz package (for visualizing error propagation)
  7. Internet Connection

GUI Dependencies:

  1. JDK7/JDK8 with JavaFX
  2. tcsh shell

Usage: run "python3 InstallLLFI.py -h" to see all running options/guidelines

  1. Copy the InstallLLFI.py script to where you want to build the LLFI
  2. Make sure you are not logged in as root
  3. Run "python3 InstallLLFI.py"
  4. Wait for compilation to finish
  5. Run the GUI by executing "./llfi-gui" under the bin/ folder

About tcsh:

The LLFI-GUI uses tcsh to read environment variables describing the location of the LLFI build. The installer will automatically add those environment variables to your ~/.tcshrc file. You do not need to actively use tcsh as your primary shell, simply having it installed is enough.

Manual Install

This method is also available, and may be more suitable if you want more control over the location of the LLVM build that the LLFI requires (ie, you already have LLVM built and wish to use that build).


  1. 64 Machine with 64 bit Linux or OS X
  2. CMake (minimum v2.8)
  3. Python 3 and above
  4. Python YAML library (PyYAML)
  5. Clang v3.4
  6. LLVM v3.4, built with CMake * Build llvm-3.4 WITH CMAKE using flag -DLLVM_REQUIRES_RTTI=1. Instructions * Remember to run make in the llvm build directory after running cmake.
  7. GraphViz package (for visualizing error propagation)

GUI Dependencies:

  1. JDK7/JDK8 with JavaFX
  2. tcsh shell


Run ./setup --help for build instructions.

  $ ./setup --help

  Usage: setup OPTIONS
  List of options:
  -LLVM_DST_ROOT <LLVM CMake build root dir>:
      Make sure you build LLVM with CMake and pass build root directory here
  -LLVM_SRC_ROOT <LLVM source root dir>
  -LLFI_BUILD_ROOT <path where you want to build LLFI>
  -LLVM_GXX_BIN_DIR <llvm-gcc/g++'s parent dir> (optional):
      You don't need to set it if it is in system path
  -JAVA_HOME_DIR <java home directory for oracle jdk 7 or higher> (optional):
    You don't need to set it if your default jdk is oracle jdk 7 or higher and in system path

  --help(-h): show help information
  --no_gui: Add this option if you do not want GUI.
  --runTests: Add this option if you want to run all regression tests after building LLFI.

Here is a sample build command if clang and javac are already in $PATH:


Build without GUI: To build LLFI without GUI, just add option: --no_gui in the command line for setup, for example:

./setup -LLFI_BUILD_ROOT $BUILD/LLFI -LLVM_SRC_ROOT $SRC/llvm-3.4 -LLVM_DST_ROOT $BUILD/llvm-3.4 --no_gui

Running tests: Running all regression tests after installation is highly recommended. Note that you may encounter some error messages during the fault injection stage. This is normal. Once all tests have completed and they all passed, LLFI is correctly installed.

VirtualBox Image

If you want to quickly try out LLFI, an Ubuntu image with LLFI and its dependencies pre-installed is available here (2.60GB). This image is built with VirtualBox v4.3.26, with Ubuntu 14.04.2 LTS, LLVM v3.4, CMake v3.4 and the current master branch version of LLFI (as of Sep 16th, 2015).

user: llfi
password: root

<LLFI_SRC_ROOT> is located under ~/Desktop/llfisrc/.
<LLFI_BUILD_ROOT> is located under ~/Desktop/llfi/.
<LLVM_SRC_ROOT> is located under ~/Desktop/llvmsrc/.
<LLVM_DST_ROOT> is located under ~/Desktop/llvm/.
<LLVM_GXX_BIN_DIR > is located under ~/Desktop/llvm/bin/.

Sample tests can be found under ~/Desktop/test/.

To run it, open VirtualBox, select File->Import Appliance... and navigate to the .ova file.


You can use test programs in the directory sample_programs/ or test_suite/PROGRAMS/ to test LLFI. Programs in the sample_programs directory already contains a valid input.yaml file. ####Command line Example program: factorial

  1. Copy the sample_programs/factorial/ directory to your project directory.
  2. Change to your factorial directory Build a single IR file with the LLFI tool GenerateMakefile
    <LLFI_BUILD_ROOT>/tools/GenerateMakefile --readable --all -o factorial.ll
    Alternatively, you can build your own IR file with clang.
  3. Instrument factorial with calls to LLFI libraries and create executables under llfi directory
    <LLFI_BUILD_ROOT>/bin/instrument --readable factorial.ll
  4. Run factorial executable with profiling functions instrumented
    <LLFI_BUILD_ROOT>/bin/profile ./llfi/factorial-profiling.exe 6
    In file llfi/baseline/golden_std_output, you should be able to see 720
  5. Run factorial executable with fault injection functions instrumented
    <LLFI_BUILD_ROOT>/bin/injectfault ./llfi/factorial-faultinjection.exe 6

For complete test of whole of LLFI, please use LLFI test suite and refer to wiki page: Test suite for regression test for details.

####GUI If you have used ./setup to install LLFI, you need to set new environment variables for tcsh shell before running the GUI for the first time. Open ~/.tcshrc using your favourite text editor and add setenv llfibuild <LLFI_BUILD_ROOT>/ and setenv zgrviewer <LLFI_BUILD_ROOT>/tools/zgrviewer/ to it. [OPTIONAL] Create an environment variable "COMPARE" with the path of the SDC check script.

Execute <LLFI_BUILD_ROOT>/bin/llfi-gui to start the GUI. The outputs will be saved in the directory where you have executed the command.

####Web GUI Development Environment Setup Dependencies: Nodejs webpack

Steps to set up the development environment:
1: Download this project from Git
2: Download NodeJs
3: Install libraries: Go to the web-app directory and run "npm install"
4: Install Webpack: In the same directory as step 3, run "sudo npm install -g webpack"
5: Configurate the LLFI root path for the server:
The default bevaiour of the program use environment variable $llfibuild as the path of the llfi build directory
You can set the environment variable llfibuild in your system to point it to the LLFI build directory in your local machine.

Start the server:
Go to the /web-app/server folder and run "node server.js"

Start the front-end dev tool:
Go to the web-app directory and run "webpack" or "webpack -w"


After fault injection, output from LLFI and the tested application can be found in the llfi directory.

Directory Contents
std_output Piped STDOUT from the tested application
llfi_stat_output Fault injection statistics
error_output Failure reports (program crashes, hangs, etc.)
trace_report_output Faults propogation report files and graph


====== Read caveats.txt for caveats and known problems.