Fuzzing the Kernel Using Unicornafl and AFL++

Overview

Unicorefuzz

Build Status code-style: black

Fuzzing the Kernel using UnicornAFL and AFL++. For details, skim through the WOOT paper or watch this talk at CCCamp19.

Is it any good?

yes.

AFL Screenshot

Unicorefuzz Setup

  • Install python2 & python3 (ucf uses python3, however qemu/unicorn needs python2 to build)
  • Run ./setup.sh, preferrably inside a Virtualenv (else python deps will be installed using --user). During install, afl++ and uDdbg as well as python deps will be pulled and installed.
  • Enjoy ucf

Upgrading

When upgrading from an early version of ucf:

  • Unicorefuzz will notify you of config changes and new options automatically.
  • Alternatively, run ucf spec to output a commented config.py spec-like element.
  • probe_wrapper.py is now ucf attach.
  • harness.py is now named ucf emu.
  • The song remains the same.

Debug Kernel Setup (Skip this if you know how this works)

  • Create a qemu-img and install your preferred OS on there through qemu
  • An easy way to get a working userspace up and running in QEMU is to follow the steps described by syzkaller, namely create-image.sh
  • For kernel customization you might want to clone your preferred kernel version and compile it on the host. This way you can also compile your own kernel modules (e.g. example_module).
  • In order to find out the address of a loaded module in the guest OS you can use cat /proc/modules to find out the base address of the module location. Use this as the offset for the function where you want to break. If you specify MODULE and BREAK_OFFSET in the config.py, it should use ./get_mod_addr.sh to start it automated.
  • You can compile the kernel with debug info. When you have compiled the linux kernel you can start gdb from the kernel folder with gdb vmlinux. After having loaded other modules you can use the lx-symbols command in gdb to load the symbols for the other modules (make sure the .ko files of the modules are in your kernel folder). This way you can just use something like break function_to_break to set breakpoints for the required functions.
  • In order to compile a custom kernel for Arch, download the current Arch kernel and set the .config to the Arch default. Then set DEBUG_KERNEL=y, DEBUG_INFO=y, GDB_SCRIPTS=y (for convenience), KASAN=y, KASAN_EXTRA=y. For convenience, we added a working example_config that can be place to the linux dir.
  • To only get necessary kernel modules boot the current system and execute lsmod > mylsmod and copy the mylsmod file to your host system into the linux kernel folder that you downloaded. Then you can use make LSMOD=mylsmod localmodconfig to only make the kernel modules that are actually needed by the guest system. Then you can compile the kernel like normal with make. Then mount the guest file system to /mnt and use make modules_install INSTALL_MOD_PATH=/mnt. At last you have to create a new initramfs, which apparently has to be done on the guest system. Here use mkinitcpio -k <folder in /lib/modules/...> -g <where to put initramfs>. Then you just need to copy that back to the host and let qemu know where your kernel and the initramfs are located.
  • Setting breakpoints anywhere else is possible. For this, set BREAKADDR in the config.py instead.
  • For fancy debugging, ucf uses uDdbg
  • Before fuzzing, run sudo ./setaflops.sh to initialize your system for fuzzing.

Run

  • ensure a target gdbserver is reachable, for example via ./startvm.sh
  • adapt config.py:
    • provide the target's gdbserver network address in the config to the probe wrapper
    • provide the target's target function to the probe wrapper and harness
    • make the harness put AFL's input to the desired memory location by adopting the place_input func config.py
    • add all EXITs
  • start ucf attach, it will (try to) connect to gdb.
  • make the target execute the target function (by using it inside the vm)
  • after the breakpoint was hit, run ucf fuzz. Make sure afl++ is in the PATH. (Use ./resumeafl.sh to resume using the same input folder)

Putting afl's input to the correct location must be coded invididually for most targets. However with modern binary analysis frameworks like IDA or Ghidra it's possible to find the desired location's address.

The following place_input method places at the data section of sk_buff in key_extract:

    # read input into param xyz here:
    rdx = uc.reg_read(UC_X86_REG_RDX)
    utils.map_page(uc, rdx) # ensure sk_buf is mapped
    bufferPtr = struct.unpack("<Q",uc.mem_read(rdx + 0xd8, 8))[0]
    utils.map_page(uc, bufferPtr) # ensure the buffer is mapped
    uc.mem_write(rdx, input) # insert afl input
    uc.mem_write(rdx + 0xc4, b"\xdc\x05") # fix tail

QEMUing the Kernel

A few general pointers. When using ./startvm.sh, the VM can be debugged via gdb. Use

$gdb
>file ./linux/vmlinux
>target remote :1234

This dynamic method makes it rather easy to find out breakpoints and that can then be fed to config.py. On top, startvm.sh will forward port 22 (ssh) to 8022 - you can use it to ssh into the VM. This makes it easier to interact with it.

Debugging

You can step through the code, starting at the breakpoint, with any given input. The fancy debugging makes use of uDdbg. To do so, run ucf emu -d $inputfile. Possible inputs to the harness (the thing wrapping afl-unicorn) that help debugging:

-d flag loads the target inside the unicorn debugger (uDdbg) -t flag enables the afl-unicorn tracer. It prints every emulated instruction, as well as displays memory accesses.

Gotchas

A few things to consider.

FS_BASE and GS_BASE

Unicorn did not offer a way to directly set model specific registers directly. The forked unicornafl version of AFL++ finally supports it. Most ugly code of earlier versions was scrapped.

Improve Fuzzing Speed

Right now, the Unicorefuzz ucf attach harness might need to be manually restarted after an amount of pages has been allocated. Allocated pages should propagate back to the forkserver parent automatically but might still get reloaded from disk for each iteration.

IO/Printthings

It's generally a good idea to nop out kprintf or kernel printing functionality if possible, when the program is loaded into the emulator.

Troubleshooting

If you got trouble running unicorefuzz, follow these rulse, worst case feel free to reach out to us, for example to @domenuk on twitter. For some notes on debugging and developing ucf and afl-unicorn further, read DEVELOPMENT.md

Just won't start

Run the harness without afl (ucf emu -t ./sometestcase). Make sure you are not in a virtualenv or in the correct one. If this works but it still crashes in AFL, set AFL_DEBUG_CHILD_OUTPUT=1 to see some harness output while fuzzing.

All testcases time out

Make sure ucf attach is running, in the same folder, and breakpoint has been triggered.

Owner
Security in Telecommunications
The Computer Security Group at Berlin University of Technology
Security in Telecommunications
Convenient tool for speeding up the intern/officer review process.

icpc-app-screen Convenient tool for speeding up the intern/officer applicant review process. Eliminates the pain from reading application responses of

1 Oct 30, 2021
NaijaSenti is an open-source sentiment and emotion corpora for four major Nigerian languages

NaijaSenti is an open-source sentiment and emotion corpora for four major Nigerian languages. This project was supported by lacuna-fund initiatives. Jump straight to one of the sections below, or jus

Hausa Natural Language Processing 14 Dec 20, 2022
Dynamic Neural Representational Decoders for High-Resolution Semantic Segmentation

Dynamic Neural Representational Decoders for High-Resolution Semantic Segmentation Requirements This repository needs mmsegmentation Training To train

Adelaide Intelligent Machines (AIM) Group 7 Sep 12, 2022
Zero-shot Learning by Generating Task-specific Adapters

Code for "Zero-shot Learning by Generating Task-specific Adapters" This is the repository containing code for "Zero-shot Learning by Generating Task-s

INK Lab @ USC 11 Dec 17, 2021
PyTorch implementation of normalizing flow models

PyTorch implementation of normalizing flow models

Vincent Stimper 242 Jan 02, 2023
[ACM MM 2021] Joint Implicit Image Function for Guided Depth Super-Resolution

Joint Implicit Image Function for Guided Depth Super-Resolution This repository contains the code for: Joint Implicit Image Function for Guided Depth

hawkey 78 Dec 27, 2022
3.8% and 18.3% on CIFAR-10 and CIFAR-100

Wide Residual Networks This code was used for experiments with Wide Residual Networks (BMVC 2016) http://arxiv.org/abs/1605.07146 by Sergey Zagoruyko

Sergey Zagoruyko 1.2k Dec 29, 2022
BABEL: Bodies, Action and Behavior with English Labels [CVPR 2021]

BABEL is a large dataset with language labels describing the actions being performed in mocap sequences. BABEL labels about 43 hours of mocap sequences from AMASS [1] with action labels.

113 Dec 28, 2022
[NeurIPS 2021] A weak-shot object detection approach by transferring semantic similarity and mask prior.

[NeurIPS 2021] A weak-shot object detection approach by transferring semantic similarity and mask prior.

BCMI 49 Jul 27, 2022
This repository contains the code and models necessary to replicate the results of paper: How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspective

Black-Box-Defense This repository contains the code and models necessary to replicate the results of our recent paper: How to Robustify Black-Box ML M

OPTML Group 2 Oct 05, 2022
Code for the paper "A Study of Face Obfuscation in ImageNet"

A Study of Face Obfuscation in ImageNet Code for the paper: A Study of Face Obfuscation in ImageNet Kaiyu Yang, Jacqueline Yau, Li Fei-Fei, Jia Deng,

35 Oct 04, 2022
Vehicle speed detection with python

Vehicle-speed-detection In the project simulate the tracker.py first then simulate the SpeedDetector.py. Finally, a new window pops up and the output

3 Dec 15, 2022
Author's PyTorch implementation of TD3+BC, a simple variant of TD3 for offline RL

A Minimalist Approach to Offline Reinforcement Learning TD3+BC is a simple approach to offline RL where only two changes are made to TD3: (1) a weight

Scott Fujimoto 193 Dec 23, 2022
X-modaler is a versatile and high-performance codebase for cross-modal analytics.

X-modaler X-modaler is a versatile and high-performance codebase for cross-modal analytics. This codebase unifies comprehensive high-quality modules i

910 Dec 28, 2022
Prior-Guided Multi-View 3D Head Reconstruction

Prior-Guided Head MVS This repository includes some reconstruction results of our IEEE TMM 2021 paper, Prior-Guided Multi-View 3D Head Reconstruction.

11 Aug 17, 2022
Code accompanying the paper "Knowledge Base Completion Meets Transfer Learning"

Knowledge Base Completion Meets Transfer Learning This code accompanies the paper Knowledge Base Completion Meets Transfer Learning published at EMNLP

14 Nov 27, 2022
Multi-tool reverse engineering collaboration solution.

CollaRE v0.3 Intorduction CollareRE is a tool for collaborative reverse engineering that aims to allow teams that do need to use more then one tool du

105 Nov 27, 2022
Material related to the Principles of Cloud Computing course.

CloudComputingCourse Material related to the Principles of Cloud Computing course. This repository comprises material that I use to teach my Principle

Aniruddha Gokhale 15 Dec 02, 2022
PyTorch implementation of paper "IBRNet: Learning Multi-View Image-Based Rendering", CVPR 2021.

IBRNet: Learning Multi-View Image-Based Rendering PyTorch implementation of paper "IBRNet: Learning Multi-View Image-Based Rendering", CVPR 2021. IBRN

Google Interns 371 Jan 03, 2023
Code for ICLR2018 paper: Improving GAN Training via Binarized Representation Entropy (BRE) Regularization - Y. Cao · W Ding · Y.C. Lui · R. Huang

code for "Improving GAN Training via Binarized Representation Entropy (BRE) Regularization" (ICLR2018 paper) paper: https://arxiv.org/abs/1805.03644 G

21 Oct 12, 2020