Fuzzer for Linux Kernel Drivers

Overview

difuze: Fuzzer for Linux Kernel Drivers

License

This repo contains all the sources (including setup scripts), you need to get difuze up and running.

Tested on

Ubuntu >= 14.04.5 LTS

0. Running difuze from Docker

Refer the readme

As explained in our paper, There are two main components of difuze: Interface Recovery and Fuzzing Engine

1. Interface Recovery

The Interface recovery mechanism is based on LLVM analysis passes. Every step of interface recovery are written as individual passes. Follow the below instructions on how to get the Interface Recovery up and running.

1.1 Setup

This step takes care of installing LLVM and c2xml:

First, make sure that you have libxml (required for c2xml):

sudo apt-get install libxml2-dev
sudo pip install lxml

Next, We have created a single script, which downloads and builds all the required tools.

cd helper_scripts
python setup_difuze.py --help
usage: setup_difuze.py [-h] [-b TARGET_BRANCH] [-o OUTPUT_FOLDER]

optional arguments:
  -h, --help        show this help message and exit
  -b TARGET_BRANCH  Branch (i.e. version) of the LLVM to setup. Default:
                    release_38 e.g., release_38
  -o OUTPUT_FOLDER  Folder where everything needs to be setup.

Example:

python setup_difuze.py -o difuze_deps

To complete the setup you also need modifications to your local PATH environment variable. The setup script will give you exact changes you need to do.

1.2 Building

This depends on the successful completion of Setup. We have a single script that builds everything, you are welcome.

cd InterfaceHandlers
./build.sh

1.3 Running

This depends on the successful completion of Build. To run the Interface Recovery components on kernel drivers, we need to first the drivers into llvm bitcode.

1.3.1 Building kernel

First, we need to have a buildable kernel. Which means you should be able to compile the kernel using regular build setup. i.e., make. We first capture the output of make command, from this output we extract the exact compilation command.

1.3.1.1 Generating output of make
Option 1: Using Bear (RECOMMENDED)
  1. Install Bear
  2. Run make using Bear:
    bear make <all the options to make>
    
    Example: bear make -j8

This will generate a file compile_commands.json in the current directory.

Option 2

Just pass V=1 and redirect the output to the file. Example:

make V=1 O=out ARCH=arm64 > makeout.txt 2>&1

NOTE: DO NOT USE MULTIPLE PROCESSES i.e., -j. Running in multi-processing mode will mess up the output file as multiple process try to write to the output file.

That's it. Next, in the following step our script takes the generated makeout.txt and run the Interface Recovery on all the recognized drivers.

1.3.2 Running Interface Recovery analysis

All the various steps of Interface Recovery are wrapped in a single script helper_scripts/run_all.py How to run:

cd helper_scripts
python run_all.py --help

usage: run_all.py [-h] [-l LLVM_BC_OUT] [-a CHIPSET_NUM] [-m MAKEOUT]
                  [-c COMPJSON] [-g COMPILER_NAME] [-n ARCH_NUM] [-o OUT]
                  [-k KERNEL_SRC_DIR] [-isclang] [-clangp CLANG_PATH]
                  [-llvmlinkp LLVMLINK_PATH] [-skb] [-skl] [-skp] [-skP]
                  [-ske] [-skI] [-ski] [-skv] [-skd] [-f IOCTL_FINDER_OUT]

optional arguments:
  -h, --help            show this help message and exit
  -l LLVM_BC_OUT        Destination directory where all the generated bitcode
                        files should be stored.
  -a CHIPSET_NUM        Chipset number. Valid chipset numbers are:
                        1(mediatek)|2(qualcomm)|3(huawei)|4(samsung)
  -m MAKEOUT            Path to the makeout.txt file.
  -c COMPJSON           Path to the compile_commands_json generated by Bear.
  -g COMPILER_NAME      Name of the compiler used in the makeout.txt, This is
                        needed to filter out compilation commands. Ex: aarch64
                        -linux-android-gcc
  -n ARCH_NUM           Destination architecture, 32 bit (1) or 64 bit (2).
  -o OUT                Path to the out folder. This is the folder, which
                        could be used as output directory during compiling
                        some kernels.
  -k KERNEL_SRC_DIR     Base directory of the kernel sources.
  -isclang              flag to indicate that clang was used to built the
                        kernel
  -clangp CLANG_PATH    Absolute path to the clang binary (if not provided,
                        the one available in the path will be used)
  -llvmlinkp LLVMLINK_PATH
                        Absolute path to the llvm-link binary (if not
                        provided, the one available in the path will be used)
  -skb                  Skip LLVM Build (default: not skipped).
  -skl                  Skip Dr Linker (default: not skipped).
  -skp                  Skip Parsing Headers (default: not skipped).
  -skP                  Skip Generating Preprocessed files (default: not
                        skipped).
  -ske                  Skip Entry point identification (default: not
                        skipped).
  -skI                  Skip Generate Includes (default: not skipped).
  -ski                  Skip IoctlCmdParser run (default: not skipped).
  -skv                  Skip V4L2 ioctl processing (default: not skipped).
  -skd                  Skip Device name finder (default: not skipped).
  -f IOCTL_FINDER_OUT   Path to the output folder where the ioctl command
                        finder output should be stored.


The script builds, links and runs Interface Recovery on all the recognized drivers, as such it might take considerable time(45 min-90 min).

The above script performs following tasks in a multiprocessor mode to make use of all CPU cores:

1.3.2.1 LLVM Build
  • Enabled by default.

All the bitcode files generated will be placed in the folder provided to the argument -l. This step takes considerable time, depending on the number of cores you have. So, if you had already done this step, You can skip this step by passing -skb.

1.3.2.2 Linking all driver bitcode files in s consolidated bitcode file.
  • Enabled by default

This performs linking, it goes through all the bitcode files and identifies the related bitcode files that need to be linked and links them (using llvm-link) in to a consolidated bitcode file (which will be stored along side corresponding bitcode file).

Similar to the above step, you can skip this step by passing -skl.

1.3.2.3 Parsing headers to identify entry function fields.
  • Enabled by default.

This step looks for the entry point declarations in the header files and stores their configuration in the file: hdr_file_config.txt under LLVM build directory.

To skip: -skp

1.3.2.4 Identify entry points in all the consolidated bitcode files.
  • Enabled by default

This step identifies all the entry points across all the driver consolidated bitcode files. The output will be stored in file: entry_point_out.txt under LLVM build directory.

Example of contents in the file entry_point_out.txt:

IOCTL:msm_lsm_ioctl:/home/difuze/kernels/pixel/msm/sound/soc/msm/qdsp6v2/msm-lsm-client.c:msm_lsm_ioctl.txt:/home/difuze/pixel/llvm_out/sound/soc/msm/qdsp6v2/llvm_link_final/final_to_check.bc
IOCTL:msm_pcm_ioctl:/home/difuze/kernels/pixel/msm/sound/soc/msm/qdsp6v2/msm-pcm-lpa-v2.c:msm_pcm_ioctl.txt:/home/difuze/pixel/llvm_out/sound/soc/msm/qdsp6v2/llvm_link_final/final_to_check.bc

To skip: -ske

1.3.2.5 Run Ioctl Cmd Finder on all the identified entry points.
  • Enabled by default.

This step will run the main Interface Recovery component (IoctlCmdParser) on all the entry points in the file entry_point_out.txt. The output for each entry point will be stored in the folder provided for option -f.

To skip: -ski

1.4 Example:

Now, we will show an example from the point where you have kernel sources to the point of getting Interface Recovery results.

We have uploaded a mediatek kernel 33.2.A.3.123.tar.bz2. First download and extract the above file.

Lets say you extracted the above file in a folder called: ~/mediatek_kernel

1.4.1 Building

Install Bear and follow the below steps:

cd ~/mediatek_kernel
source ./env.sh
cd kernel-3.18
# the following step may not be needed depending on the kernel
mkdir out
make O=out ARCH=arm64 tubads_defconfig
# generating compile_commands.json
bear make -j8 O=out ARCH=arm64

1.4.2 Running Interface Recovery

cd <repo_path>/helper_scripts

python run_all.py -l ~/mediatek_kernel/llvm_bitcode_out -a 1 -c ~/mediatek_kernel/kernel-3.18/compile_commands.json -n 2 -o ~/mediatek_kernel/kernel-3.18/out -k ~/mediatek_kernel/kernel-3.18 -f ~/mediatek_kernel/ioctl_finder_out

The above command takes quite some time (30 min - 1hr).

1.4.3 Understanding the output

First, all the analysis results will be in the folder: ~/mediatek_kernel/ioctl_finder_out (argument given to the option -f), for each entry point a .txt file will be created, which contains all the information about the recovered interface.

If you are interested in information about just the interface and don't care about anything else, We recommend you use the parse_interface_output.py script. This script converts the crazy output of Interface Recovery pass into nice json files with a clean and consistent format.

cd <repo_path>/helper_scripts
python parse_interface_output.py <ioctl_finder_out_dir> <output_directory_for_json_files>

Here <ioctl_finder_out_dir> should be same as the folder you provided to the -f option and <output_directory_for_json_files> is the folder where the json files should be created.

You can use the corresponding json files for the interface recovery of the corresponding ioctl.

1.4.4 Things to note:

1.4.4.1 Value for option -g (only if you use makeout.txt)

To provide value for option -g you need to know the name of the *-gcc binary used to compile the kernel. An easy way to know this would be to grep for gcc in makeout.txt and you will see compiler commands from which you can know the *-gcc binary name.

For our example above, if you do grep gcc makeout.txt for the example build, you will see lot of lines like below:

aarch64-linux-android-gcc -Wp,-MD,fs/jbd2/.transaction.o.d  -nostdinc -isystem ...

So, the value for -g should be aarch64-linux-android-gcc.

If the kernel to be built is 32-bit then the binary most likely will be arm-eabi-gcc

For Qualcomm (or msm) chipsets, you may see *gcc-wrapper.py instead of *.gcc, in which case you should provide the *gcc-wrapper.py.

1.4.4.2 Value for option -a

Depeding on the chipset type, you need to provide corresponding number.

1.4.4.3 Value for option -o

This is the path of the folder provided to the option O= for make command during kernel build.

Not all kernels need a separate out path. You may build kernel by not providing an option O, in which case you SHOULD NOT provide value for that option while running run_all.py.

Kernels built using clang

For kernels built using clang, in addition to the above options please specify the following options (assuming you used compile_commands.json):

-isclang -clangp <PATH_TO_THE_CLANG_USED_TO_BUILD_THE_KERNEL> -llvmlinkp <PATH_TO_THE_LLVM_LINK (will be in the same folder as clang)>

1.5 Post Processing

Before we can begin fuzzing we need to process the output a bit with our very much research quality (sorry) parsers.

These are found here. The main script to run will be run_all.py:

$ python run_all.py --help
usage: run_all.py [-h] -f F -o O [-n {manual,auto,hybrid}] [-m M]

run_all options

optional arguments:
  -h, --help            show this help message and exit
  -f F                  Filename of the ioctl analysis output OR the entire
                        output directory created by the system
  -o O                  Output directory to store the results. If this
                        directory does not exist it will be created
  -n {manual,auto,hybrid}
                        Specify devname options. You can choose manual
                        (specify every name manually), auto (skip anything that
                        we don't identify a name for), or hybrid (if we
                        detected a name, we use it, else we ask the user)
  -m M                  Enable multi-device output most ioctls only have one
                        applicable device node, but some may have multiple. (0
                        to disable)

You'll want to pass -f the output directory of the ioctl analysis e.g. ~/mediatek_kernel/ioctl_finder_out.

-o Is where you where to store the post-processed results. These will be easily digestible XML files (jpits).

-n Specifies the system to what degree you want to rely on our device name recovery. If you don't want to do any work/name hunting, you can specify auto. This of course comes at the cost of skipping any device for which we don't recover a name. If you want to be paranoid and not trust any of our recovery efforts (totally reasonable) you can use the manual option to name every single device yourself. hybrid then is a combination of both -- we will name the device for you when we can, and fall back to you when we've failed.

-m Sometimes ioctls can correspond to more than one device (this is common with v4l2/subdev ioctls for example). Support for this in enabled by default, but it requires user interaction to specify the numberof devices for each device. If this is too annoying for you, you can disable the prompt by passing -m 0 (we will assume a single device for each ioctl).

After running, you should have, in your out folder, a folder for each ioctl.

2 Fuzzing

2.1 Mango Fuzz

MangoFuzz is our simple prototype fuzzer and is based off of Peach (specifically MozPeach).

It's not a particularly sophisticated fuzzer but it does find bugs. It was also built to be easily expandable. There are 2 components to this fuzzer, the fuzz engine and the executor. The executor can be found here, and the fuzz engine can be found here.

2.1.1 Executor

The executor runs on the phone, listening for data that the fuzz engine will send to it.

Simply compile it for your phones architecture, adb push it on to the phone, and execute with the port you want it to listen on!

2.1.2 Fuzz Engine

Interfacing with MangoFuzz is fairly simple. You'll want an Engine object and a Parser object, which you'll feed your engine into. From here, you parse jpits with your Parser, and then run the Engine. Easy! We've provided some simple run scripts to get you started.

To run against specific drivers you can use runner.py on one of the ioctl folders in the output directory (created by our post processing scripts).

e.g. ./runner.py -f honor8/out/chb -num 1000. This tells MangoFuzz to run for 1000 iterations against all ioctl command value pairs pertaining to the chb ioctl/driver.

If instead we want to run against an entire device (phone), you can use dev_runner.py. e.g. ./dev_runner.py -f honor8/out -num 100. This will continue looping over the driver files, randomly switching between them for 100 iterations each.

Note that before the fuzz engine can communicate with the phone, you'll need to use ADB to set up port forwarding e.g. adb forward tcp:2022 tcp:2022

Owner
seclab
The Computer Security Group at UC Santa Barbara
seclab
MPI-IS Mesh Processing Library

Perceiving Systems Mesh Package This package contains core functions for manipulating meshes and visualizing them. It requires Python 3.5+ and is supp

Max Planck Institute for Intelligent Systems 494 Jan 06, 2023
Kaggle | 9th place (part of) solution for the Bristol-Myers Squibb – Molecular Translation challenge

Part of the 9th place solution for the Bristol-Myers Squibb – Molecular Translation challenge translating images containing chemical structures into I

Erdene-Ochir Tuguldur 22 Nov 30, 2022
Fine-grained Control of Image Caption Generation with Abstract Scene Graphs

Faster R-CNN pretrained on VisualGenome This repository modifies maskrcnn-benchmark for object detection and attribute prediction on VisualGenome data

Shizhe Chen 7 Apr 20, 2021
RepMLP: Re-parameterizing Convolutions into Fully-connected Layers for Image Recognition

RepMLP: Re-parameterizing Convolutions into Fully-connected Layers for Image Recognition (PyTorch) Paper: https://arxiv.org/abs/2105.01883 Citation: @

260 Jan 03, 2023
Code of paper "Compositionally Generalizable 3D Structure Prediction"

Compositionally Generalizable 3D Structure Prediction In this work, We bring in the concept of compositional generalizability and factorizes the 3D sh

Songfang Han 30 Dec 17, 2022
🏖 Keras Implementation of Painting outside the box

Keras implementation of Image OutPainting This is an implementation of Painting Outside the Box: Image Outpainting paper from Standford University. So

Bendang 1.1k Dec 10, 2022
Pytorch implementation of CVPR2020 paper “VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation”

VectorNet Re-implementation This is the unofficial pytorch implementation of CVPR2020 paper "VectorNet: Encoding HD Maps and Agent Dynamics from Vecto

120 Jan 06, 2023
Pointer networks Tensorflow2

Pointer networks Tensorflow2 原文:https://arxiv.org/abs/1506.03134 仅供参考与学习,内含代码备注 环境 tensorflow==2.6.0 tqdm matplotlib numpy 《pointer networks》阅读笔记 应用场景

HUANG HAO 7 Oct 27, 2022
Repository containing detailed experiments related to the paper "Memotion Analysis through the Lens of Joint Embedding".

Memotion Analysis Through The Lens Of Joint Embedding This repository contains the experiments conducted as described in the paper 'Memotion Analysis

Nethra Gunti 1 Mar 16, 2022
[ACMMM 2021, Oral] Code release for "Elastic Tactile Simulation Towards Tactile-Visual Perception"

EIP: Elastic Interaction of Particles Code release for "Elastic Tactile Simulation Towards Tactile-Visual Perception", in ACMMM (Oral) 2021. By Yikai

Yikai Wang 37 Dec 20, 2022
HIVE: Evaluating the Human Interpretability of Visual Explanations

HIVE: Evaluating the Human Interpretability of Visual Explanations Project Page | Paper This repo provides the code for HIVE, a human evaluation frame

Princeton Visual AI Lab 16 Dec 13, 2022
Python implementation of the multistate Bennett acceptance ratio (MBAR)

pymbar Python implementation of the multistate Bennett acceptance ratio (MBAR) method for estimating expectations and free energy differences from equ

Chodera lab // Memorial Sloan Kettering Cancer Center 169 Dec 02, 2022
D2Go is a toolkit for efficient deep learning

D2Go D2Go is a production ready software system from FacebookResearch, which supports end-to-end model training and deployment for mobile platforms. W

Facebook Research 744 Jan 04, 2023
Geometry-Free View Synthesis: Transformers and no 3D Priors

Geometry-Free View Synthesis: Transformers and no 3D Priors Geometry-Free View Synthesis: Transformers and no 3D Priors Robin Rombach*, Patrick Esser*

CompVis Heidelberg 293 Dec 22, 2022
Disturbing Target Values for Neural Network regularization: attacking the loss layer to prevent overfitting

Disturbing Target Values for Neural Network regularization: attacking the loss layer to prevent overfitting 1. Classification Task PyTorch implementat

Yongho Kim 0 Apr 24, 2022
FLAVR is a fast, flow-free frame interpolation method capable of single shot multi-frame prediction

FLAVR is a fast, flow-free frame interpolation method capable of single shot multi-frame prediction. It uses a customized encoder decoder architecture with spatio-temporal convolutions and channel ga

Tarun K 280 Dec 23, 2022
PyTorch implementation of the Crafting Better Contrastive Views for Siamese Representation Learning

Crafting Better Contrastive Views for Siamese Representation Learning This is the official PyTorch implementation of the ContrastiveCrop paper: @artic

249 Dec 28, 2022
Official repository for the CVPR 2021 paper "Learning Feature Aggregation for Deep 3D Morphable Models"

Deep3DMM Official repository for the CVPR 2021 paper Learning Feature Aggregation for Deep 3D Morphable Models. Requirements This code is tested on Py

38 Dec 27, 2022
Official implementation of the NeurIPS 2021 paper Online Learning Of Neural Computations From Sparse Temporal Feedback

Online Learning Of Neural Computations From Sparse Temporal Feedback This repository is the official implementation of the NeurIPS 2021 paper Online L

Lukas Braun 3 Dec 15, 2021
Semi-supervised Stance Detection of Tweets Via Distant Network Supervision

SANDS This is an annonymous repository containing code and data necessary to reproduce the results published in "Semi-supervised Stance Detection of T

2 Sep 22, 2022