Skip to content

Latest commit

 

History

History
278 lines (214 loc) · 16.1 KB

README.md

File metadata and controls

278 lines (214 loc) · 16.1 KB

MUDscope

This repository contains the code for MUDscope by the authors of the ACSAC 2022 paper "Stepping out of the MUD: Contextual threat information for IoT devices with manufacturer-provided behaviour profiles" PDF [1]. Please cite MUDscope when using it in academic publications.

Table of Contents

Introduction

Besides coming with unprecedented benefits, the Internet of Things (IoT) suffers deficits in security measures, leading to attacks increas- ing every year. In particular, network environments such as smart homes lack managed security capabilities to detect IoT-related at- tacks; IoT devices hosted therein are thus more easily infiltrated by threats. As such, context awareness on IoT infections is hard to achieve, preventing prompt response. In this work, we propose MUDscope, an approach to monitor malicious network activities affecting IoT in real-world consumer environments. We leverage the recent Manufacturer Usage Description (MUD) specification, which defines networking whitelists for IoT devices in MUD pro- files, to reflect consistent and necessarily-anomalous activities from smart things. Our approach characterizes this traffic and extracts signatures for given attacks. By analyzing attack signatures for multiple devices, we gather insights into emerging attack patterns. We evaluate our approach on both an existing dataset, and a new openly available dataset created for this research. We show that MUDscope detects several attacks targeting IoT devices with an F1-score of 95.77% and correctly identifies signatures for specific attacks with an F1-score of 87.72%.

Overview

The goal of MUDscope is to provide a distributed Network Telescope for IoT-related traffic that uses MUD enforcers as a specification-based IDS to consistently detect malicious network traffic across deployments. To this end, we take the following steps:

  1. Create MUD profiles for each monitored device. One can use MUD profiles provided by the IoT manufacturer. In this work, we automatically generate MUD profiles from benign IoT network traffic using MUDgee.
  2. Filter IoT network traffic (pcap files) based on MUD profile, called MUD-rejected traffic (MRT).
  3. Group filtered network traffic into NetFlows per device per time window.
  4. Cluster NetFlows using HDBSCAN to obtain groups of similar traffic (characterisations).
  5. Compare characterisations of subsequent time windows to describe the evolution of MRT over time.
  6. Compare MRT descriptions from multiple devices to provide insights on how anomalous activities affect the selected devices.

Installation

The easiest way of installing MUDscope including all its dependencies is using our Docker container:

git clone https://github.com/lucamrgs/MUDscope.git  # Clone repository
cd MUDscope                                         # Change into MUDscope directory
docker build .                                      # Build docker container
docker container run -it <image> bash               # Run bash from within the built docker container, see Usage/examples for using MUDscope

Manual installation

Alternatively, you can install MUDscope manually:

  1. Clone our repository
git clone [email protected]:lucamrgs/MUDscope.git     # Using SSH
git clone https://github.com/lucamrgs/MUDscope.git # Using HTTPS
  1. Make sure you have installed all Dependencies
  2. Install MUDscope as python tool:
pip3 install -e /path/to/directory/containing/mudscope/pyproject.toml/

Dependencies

Please install the following dependencies to run MUDscope

# Dependencies
sudo apt install libpcap-dev

# Installation
git clone https://github.com/phaag/nfdump.git # Clone nfdump directory
cd nfdump                                     # Change to nfdump directory
git checkout v1.6.24                          # Ensure we are working with v1.6.24
./autogen.sh                                  # Setup configuration
./configure --enable-nfpcapd                  # Build nfpcapd
make                                          # Make file
make install                                  # Make file
sudo cp ./bin/.libs/libnfdump-1.6.24.so /usr/lib/ # Because library does not install correctly

When all dependencies have been installed, make sure you have the correct python libraries installed by installing them from the requirements.txt file:

pip3 install -r requirements.txt

Usage

Once MUDscope is installed, you can use the tool to execute the steps described in our paper. To this end, MUDscope works in different modes corresponding to the steps of our approach:

python3 -m mudscope <mode> [-h]

Which supports the following modes:

  1. mudgen, creates MUD files from benign network traces.
  2. reject, filters MUD rejected traffic from pcap files.
  3. netflows, transforms MRT pcap files to NetFlows.
  4. characterize, Perform characterization analysis on MRT netflows.
  5. evolution, Perform evolution analysis on characterization files to produce MRT feeds.
  6. monitor, Monitor and compare anomalous activities captured in MRT feeds.
usage: __main__.py [-h] {mudgen,reject,netflows,characterize,evolution,monitor} ...

MUDscope - Stepping out of the MUD: Contextual threat information for IoT devices with manufacturer-
provided behaviour profiles.

optional arguments:
  -h, --help            show this help message and exit

mode:
  Mode in which to run MUDscope.

  {mudgen,reject,netflows,characterize,evolution,monitor}
    mudgen              Create MUD files from benign network traces.
    reject              Filter MUD rejected traffic from pcap files.
    netflows            Transform MRT pcap files to NetFlows.
    characterize        Perform characterization analysis on MRT netflows.
    evolution           Perform evolution analysis on characterization files to produce MRT feeds.
    monitor             Monitor and compare anomalous activities captured in MRT feeds.

mudgen

MUDscope enforces MUD profiles in pcap files to filter malicious traffic from these files. However, manufacturers often do not specify MUD profiles for their IoT devices. Therefore, we provide an easy interface to learn MUD profiles from a trace of benign network traffic of a device using the tool MUDgee. This mode takes a MUDgee config file as input and produces MUD profiles for network traces specified in this config file. See MUDgee or one of our examples for the format of these config files.

usage: __main__.py mudgen [-h] --config <path>

Create MUD files from benign network traces.

optional arguments:
  -h, --help       show this help message and exit
  --config <path>  path to JSON config file for mudgee MUD generation

Note: by default the MUD profiles will be stored in a (newly created) result directory, which is the default behaviour of MUDgee.

reject

MUDscope enforces MUD profiles by taking configs specifying pcap files for which to filter given MUD rules. When enforcing MUD profiles, traffic that does not conform to the MUD specification will be rejected and stored as a separate pcap file in the specified output directory.

usage: __main__.py reject [-h] --config <path> [<path> ...] --rules <path> --output <path>
                          [--limit <int>]

Filter MUD rejected traffic from pcap files.

optional arguments:
  -h, --help            show this help message and exit
  --config <path> [<path> ...]
                        path(s) to JSON config file(s) for MUD enforcement
  --rules <path>        path to MUD rules CSV file generated by MUDgee
  --output <path>       path to output directory in which to store results
  --limit <int>         optional, limits the number of packets processed when rejecting traffic

Note: We provide an example for the format of accepted config files. However, to create these reject config files manually, you can use the script:

python3 -m mudscope.generate_rjt_config

We recommend using this script instead of manual configuration as it automatically generates multiple reject config files referred to multiple (time-window) pcap files located in a directory, for instance the one used with editcap. See script code directly, or run python3 -m mudscope.generate_rjt_config -h to consult usage.

netflows

The pcap files containing MUD-rejected traffic (MRT) must be transformed into NetFlows which are used by the remainder of MUDscope. To transform MRT pcap files into NetFlow files, we use MUDscope's netflows mode that takes as input a directory containing all MRT pcap files, transforms them into NetFlow files that will be stored in the output directory.

usage: __main__.py netflows [-h] --input <path> --output <path>

Transform MRT pcap files to NetFlows.

optional arguments:
  -h, --help       show this help message and exit
  --input <path>   path to directory containing MUD-rejected pcap files
  --output <path>  path to output directory in which to store NetFlows

Note: MUDscope expects that input pcaps are already into separate time-windows, i.e. each pcap represents a new time-window. These separate pcaps are required during the netflows, characterize, and evolution modes. Therefore, as input for this step, make sure that your malicious pcap files are split into time-windows. You can divide pcap files spanning multiple time-windows into smaller pcaps using editcap:

editcap -i <60> path/to/input.pcap path/to/output.pcap

This splits a pcap file into smaller files each containing traffic for -i seconds, outputting all generated files to the specified path. See editcap doc for more information.

In our running example, this split has already been performed.

characterize

Using the generated NetFlows, MUDscope clusters the traffic for multiple time windows and creates characterization files describing these clusters. In our paper, we show that these clusters often correspond to different types of attacks. To generate these characterization files, MUDscope takes as input the paths to the CSV files containing MRT netflows generated in the previous step. It also requires metadata containing information about the device we are characterizing (see our examples for the format), and it requires a dsr (Dataset Scaler Reference CSV file, see examples) to perform correct feature scaling. Using these inputs, MUDscope outputs characterization files in the given output directory.

usage: __main__.py characterize [-h] --input <path> [<path> ...] --metadata <path> --dsr <path>
                                --output <path>

Perform characterization analysis on MRT netflows.

optional arguments:
  -h, --help            show this help message and exit
  --input <path> [<path> ...]
                        path(s) to CSV file(s) containing MRT netflows
  --metadata <path>     path to JSON file describing the capture to analyse
  --dsr <path>          path to Dataset Scaler Reference (DSR) CSV file
  --output <path>       output directory in which to store analyzed file(s)

Note: When characterizing traffic, MUDscope translates pcaps to bi-directional flows, and scales their features. To do this, a dataset scaling reference (DSR) is required. A DSR file can be created from a single pcap file with normal (benign) network activity. It does not need to be deployment specific, and there are no length requirements. A 1 hour-long capture obtained from sniffing the network traffic of the deployment containing your IoT devices should work well. Here, longer and more exaustive captures will provide a better scaling reference.

To generate a DSR yourself, you can use the following scripts, also see example:

# Transform benign pcaps into netflows
python3 -m mudscope.device_mrt_pcaps_to_csv \
    <path/to/directory/containing/benign/pcap/files/> \
    --outdir <path/to/output/directory/>

# Create dataset scaling reference (DSR)
python3 -m mudscope.scale_reference_df_script \
    <path/to/output/csv/of/previous/step.csv>

evolution

We now have produced MUDscope characterizations of malicious traffic for each time window. However, we would also like to analyze how these characterizations evolve over time in order to produce MRT feeds. MUDscope's evolution mode takes as input the paths to JSON characterization files, analyzes the evolution of these characterizations and stores them in the MRT feed format in the specified output file.

usage: __main__.py evolution [-h] --input <path> [<path> ...] --output <path>

Perform evolution analysis on characterization files to produce MRT feeds.

optional arguments:
  -h, --help            show this help message and exit
  --input <path> [<path> ...]
                        path(s) to file(s) containing JSON characterization files
  --output <path>       output file in which to store MRT feed

monitor

After generating MRT feeds for each device, we can compare feeds from different devices or vendors to see larger trends in the attacks targeting these devices. To this end, MUDscope provides the monitor mode which takes as input a config file specifying which MRT feeds to analyze and for which features to analyze them (see example for an example config file). Next, it will output its report and produced plots into the specified output directory.

usage: __main__.py monitor [-h] --config <path> --output <path>

Monitor and compare anomalous activities captured in MRT feeds.

optional arguments:
  -h, --help       show this help message and exit
  --config <path>  monitor config file specifying MRT feeds to compare
  --output <path>  path to directory which to write output monitor plots

Examples

We provide a running example in the examples/ directory.

We provide the complete results set of our research at: https://mega.nz/folder/hx8VgRxa#9tBD8Mh8DplIzfobQcF45w

Dataset

Our evaluations were performed on two datasets:

References

[1] Luca Morgese Zangrandi, Thijs van Ede, Tim Booij, Savio Sciancalepore, Luca Allodi, and Andrea Continella. 2022. Stepping out of the MUD: Contextual threat information for IoT devices with manufacturer-provided behaviour profiles. In Proceedings of ACSAC ’22: ACM Annual Computer Security Applications Conference (ACSAC ’22).

Bibtex

@inproceedings{morgesezangrandi2022mudscope,
  title={{Stepping out of the MUD: Contextual threat information for IoT devices with manufacturer-provided behavior profiles}},
  author={Morgese Zangrandi, Luca and van Ede, Thijs and Booij, Tim and Sciancalepore, Savio and Allodi, Luca and Continella, Andrea},
  booktitle={Proceedings of ACSAC '22: ACM Annual Computer Security Applications Conference (ACSAC '22).},
  year={2022},
  organization={ACM}
}