Geneva is an artificial intelligence tool that defeats censorship by exploiting bugs in censors

Overview

Geneva Build Status codecov Documentation Status

Are you using Geneva? If so, let us know! Shoot us an email at [email protected], or to use PGP, email us directly with our keys on our website.

Geneva is an artificial intelligence tool that defeats censorship by exploiting bugs in censors, such as those in China, India, and Kazakhstan. Unlike many other anti-censorship solutions which require assistance from outside the censoring regime (Tor, VPNs, etc.), Geneva runs strictly on one side of the connection (either the client or server side).

Under the hood, Geneva uses a genetic algorithm to evolve censorship evasion strategies and has found several previously unknown bugs in censors. Geneva's strategies manipulate the network stream to confuse the censor without impacting the client/server communication. This makes Geneva effective against many types of in-network censorship (though it cannot be used against IP-blocking censorship).

Geneva is composed of two high level components: its genetic algorithm (which it uses to evolve new censorship evasion strategies) and its strategy engine (which is uses to run an individual censorship evasion strategy over a network connection).

This codebase contains the Geneva's full implementation: its genetic algorithm, strategy engine, Python API, and a subset of published strategies. With these tools, users and researchers alike can evolve new strategies or leverage existing strategies to evade censorship. To learn more about how Geneva works, see How it Works or checkout our documentation.

Setup

Geneva has been developed and tested for Centos or Debian-based systems. Due to limitations of netfilter and raw sockets, Geneva does not work on OS X or Windows at this time and requires python3.6. More detailed setup instructions are available at our documentation.

Install netfilterqueue dependencies:

# sudo apt-get install build-essential python-dev libnetfilter-queue-dev libffi-dev libssl-dev iptables python3-pip

Install Python dependencies:

# python3 -m pip install -r requirements.txt

On Debian 10 systems, some users have reported needing to install netfilterqueue directly from Github:

# sudo python3 -m pip install --upgrade -U git+https://github.com/kti/python-netfilterqueue

Running a Strategy

A censorship evasion strategy is simply a description of how network traffic should be modified. A strategy is not code, it is a description that tells the engine how it should operate over traffic. For a fuller description of the DNA syntax, see Censorship Evasion Strategies.

# python3 engine.py --server-port 80 --strategy "[TCP:flags:PA]-duplicate(tamper{TCP:dataofs:replace:10}(tamper{TCP:chksum:corrupt},),)-|" --log debug
2019-10-14 16:34:45 DEBUG:[ENGINE] Engine created with strategy \/ (ID bm3kdw3r) to port 80
2019-10-14 16:34:45 DEBUG:[ENGINE] Configuring iptables rules
2019-10-14 16:34:45 DEBUG:[ENGINE] iptables -A OUTPUT -p tcp --sport 80 -j NFQUEUE --queue-num 1
2019-10-14 16:34:45 DEBUG:[ENGINE] iptables -A INPUT -p tcp --dport 80 -j NFQUEUE --queue-num 2
2019-10-14 16:34:45 DEBUG:[ENGINE] iptables -A OUTPUT -p udp --sport 80 -j NFQUEUE --queue-num 1
2019-10-14 16:34:45 DEBUG:[ENGINE] iptables -A INPUT -p udp --dport 80 -j NFQUEUE --queue-num 2

Note that if you have stale iptables rules or other rules that rely on Geneva's default queues, this will fail. To fix this, remove those rules.

Also note that if you want to specify multiple ports for Geneva to monitor, you can specify a port range using --server-port 4000:5000 to monitor all ports in the range 4000-5000, or you can specify a list like --server-port 80,443,4444 to only monitor the explicit ports given.

Strategy Library

Geneva has found dozens of strategies that work against censors in China, Kazakhstan, India, and Iran. We include several of these strategies in strategies.md. Note that this file contains success rates for each individual country; a strategy that works in one country may not work as well as other countries.

Researchers have observed that strategies may have differing success rates based on your exact location. Although we have not observed this from our vantage points, you may find that some strategies may work differently in a country we have tested. If this is the case, don't be alarmed. However, please feel free to reach out to a member of the team directly or open an issue on this page so we can track how the strategies work from other geographic locations.

Disclaimer

Running these strategies may place you at risk if you use it within a censoring regime. Geneva takes overt actions that interfere with the normal operations of a censor and its strategies are detectable on the network. During the training process, Geneva will intentionally trip censorship many times. Geneva is not an anonymity tool, nor does it encrypt any traffic. Understand the risks of running Geneva in your country before trying it.


How it Works

See our paper for an in-depth read on how Geneva works. Below is a walkthrough of the main concepts behind Geneva, the major components of the codebase, and how they can be used.

Censorship Evasion Strategies

A censorship evasion strategy is simply a description of how network traffic should be modified. A strategy is not code, it is a description that tells Geneva's stratgy engine how it should manipulate network traffic. The goal of a censorship evasion strategy is to modify the network traffic in a such a way that the censor is unable to censor it, but the client/server communication is unimpacted.

A censorship evasion strategy composed of one or more packet-level building blocks. Geneva's core building blocks are:

  1. duplicate: takes one packet and returns two copies of the packet
  2. drop: takes one packet and returns no packets (drops the packet)
  3. tamper: takes one packet and returns the modified packet
  4. fragment: takes one packet and returns two fragments or two segments

Since duplicate and fragment introduce branching, these actions are composed into a binary-tree structure called an action tree. Each tree also has a trigger. The trigger describes which packets the tree should run on, and the tree describes what should happen to each of those packets when the trigger fires. Once a trigger fires on a packet, it pulls the packet into the tree for modifications, and the packets that emerge from the tree are sent on the wire. Recall that Geneva operates at the packet level, therefore all triggers are packet-level triggers.

Multiple action trees together form a forest. Geneva handles outbound and inbound packets differently, so strategies are composed of two forests: an outbound forest and an inbound forest.

Consider the following example of a simple Geneva strategy.

                   +---------------+
                   |  TCP:flags:A  |         <-- triggers on TCP packets with the flags field set to 'ACK'
                   +-------+-------+             matching packets are captured and pulled into the tree
                           |
                 +---------v---------+
                       duplicate             <-- makes two copies of the given packet. the tree is processed 
                 +---------+---------+           with an inorder traversal, so the left side is run first
                           |
             +-------------+------------+
             |                          |
+------------v----------+               v    <-- dupilcate has no right child, so this packet will be sent on the wire unimpacted
          tamper              
  {TCP:flags:replace:R}      <-- parameters to this action describe how the packet should be tampered 
+------------+----------+
             |
+------------v----------+
          tamper
   {TCP:chksum:corrupt}
+------------+----------+
             |
             v               <-- packets that emerge from an in-order traversal of the leaves are sent on the wire

This strategy triggers on TCP packets with the flags field set to ACK. It makes a duplicate of the ACK packet; the first duplicate has its flags field changed to RST and its checksum (chksum) field corrupted; the second duplicate is unchaged. Both packets are then sent on the network.

Strategy DNA

These strategies can be arbitrarily complicated, and Geneva defines a well-formatted string syntax for unambiguously expressing strategies.

A strategy divides how it handles outbound and inbound packets: these are separated in the DNA by a "\/". Specifically, the strategy format is \/ . If \/ is not present in a strategy, all of the action trees are in the outbound forest.

Both forests are composed of action trees, and each forest is allowed an arbitrarily many trees.

Action trees always start with a trigger, which is formatted as: [ : : ] . For example, the trigger: [TCP:flags:S] will run its corresponding tree whenever it sees a TCP packet with the flags field set to SYN. If the corresponding action tree is [TCP:flags:S]-drop-|, this action tree will cause the engine to drop any SYN packets. [TCP:flags:S]-duplicate-| will cause the engine to duplicate any SYN packets.

Syntactically, action trees end with -|.

Depending on the type of action, some actions can have up to two children (such as duplicate). These are represented with the following syntax: [TCP:flags:S]-duplicate( , )-| , where and themselves are trees. If (,) is not specified, any packets that emerge from the action will be sent on the wire. If an action only has one child (such as tamper), it is always the left child. [TCP:flags:S]-tamper{ }( ,)-|

Actions that have parameters specify those parameters within {}. For example, giving parameters to the tamper action could look like: [TCP:flags:S]-tamper{TCP:flags:replace:A}-|. This strategy would trigger on TCP SYN packets and replace the TCP flags field to ACK.

Putting this all together, below is the strategy DNA representation of the above diagram:

[TCP:flags:A]-duplicate(tamper{TCP:flags:replace:R}(tamper{TCP:chksum:corrupt},),)-| \/

Geneva has code to parse this strategy DNA into strategies that can be applied to network traffic using the engine.

Note that due to limitations of Scapy and NFQueue, actions that introduce branching (fragment, duplicate) are disabled for incoming action forests.

Engine

The strategy engine (engine.py) applies a strategy to a network connection. The engine works by capturing all traffic to/from a specified port. Packets that match an active trigger are run through the associated action-tree, and packets that emerge from the tree are sent on the wire.

The engine also has a Python API for using it in your application. It can be used as a context manager or invoked in the background as a thread. For example, consider the following simple application.

import os
import engine

# Port to run the engine on
port = 80
# Strategy to use
strategy = "[TCP:flags:A]-duplicate(tamper{TCP:flags:replace:R}(tamper{TCP:chksum:corrupt},),)-| \/"

# Create the engine in debug mode
with engine.Engine(port, strategy, log_level="debug") as eng:
    os.system("curl http://example.com?q=ultrasurf")

This script creates an instance of the engine with a specified strategy, and that strategy will be running for everything within the context manager. When the context manager exits, the engine will clean itself up. See the examples/ folder for more use cases of the engine.

Due to limitations of scapy and NFQueue, the engine cannot be used to communicate with localhost.

Citation

If you like the work or plan to use it in your projects, please follow the guidelines in citation.bib.

Paper

See our paper from CCS or the rest of our papers and talks for an in-depth dive into how Geneva works and how it can be applied.

Contributors

Kevin Bock

George Hughey

Xiao Qiang

Dave Levin

Pytorch implementation of NeurIPS 2021 paper: Geometry Processing with Neural Fields.

Geometry Processing with Neural Fields Pytorch implementation for the NeurIPS 2021 paper: Geometry Processing with Neural Fields Guandao Yang, Serge B

Guandao Yang 162 Dec 16, 2022
Official repository of the paper 'Essentials for Class Incremental Learning'

Essentials for Class Incremental Learning Official repository of the paper 'Essentials for Class Incremental Learning' This Pytorch repository contain

33 Nov 27, 2022
Implementation of "GNNAutoScale: Scalable and Expressive Graph Neural Networks via Historical Embeddings" in PyTorch

PyGAS: Auto-Scaling GNNs in PyG PyGAS is the practical realization of our G NN A uto S cale (GAS) framework, which scales arbitrary message-passing GN

Matthias Fey 139 Dec 25, 2022
This is the official repository of the paper Stocastic bandits with groups of similar arms (NeurIPS 2021). It contains the code that was used to compute the figures and experiments of the paper.

Experiments How to reproduce experimental results of Stochastic bandits with groups of similar arms submitted paper ? Section 5 of the paper To reprod

Fabien 0 Oct 25, 2021
Controlling a game using mediapipe hand tracking

These scripts use the Google mediapipe hand tracking solution in combination with a webcam in order to send game instructions to a racing game. It features 2 methods of control

3 May 17, 2022
MultiTaskLearning - Multi Task Learning for 3D segmentation

Multi Task Learning for 3D segmentation Perception stack of an Autonomous Drivin

2 Sep 22, 2022
Minimisation of a negative log likelihood fit to extract the lifetime of the D^0 meson (MNLL2ELDM)

Minimisation of a negative log likelihood fit to extract the lifetime of the D^0 meson (MNLL2ELDM) Introduction The average lifetime of the $D^{0}$ me

Son Gyo Jung 1 Dec 17, 2021
Learning Spatio-Temporal Transformer for Visual Tracking

STARK The official implementation of the paper Learning Spatio-Temporal Transformer for Visual Tracking Hiring research interns for visual transformer

Multimedia Research 484 Dec 29, 2022
PPO Lagrangian in JAX

PPO Lagrangian in JAX This repository implements PPO in JAX. Implementation is tested on the safety-gym benchmark. Usage Install dependencies using th

Karush Suri 2 Sep 14, 2022
Generic Foreground Segmentation in Images

Pixel Objectness The following repository contains pretrained model for pixel objectness. Please visit our project page for the paper and visual resul

Suyog Jain 157 Nov 21, 2022
《Image2Reverb: Cross-Modal Reverb Impulse Response Synthesis》(2021)

Image2Reverb Image2Reverb is an end-to-end neural network that generates plausible audio impulse responses from single images of acoustic environments

Nikhil Singh 48 Nov 27, 2022
An original implementation of "Noisy Channel Language Model Prompting for Few-Shot Text Classification"

Channel LM Prompting (and beyond) This includes an original implementation of Sewon Min, Mike Lewis, Hannaneh Hajishirzi, Luke Zettlemoyer. "Noisy Cha

Sewon Min 92 Jan 07, 2023
kullanışlı ve işinizi kolaylaştıracak bir araç

Hey merhaba! işte çok sorulan sorularının cevabı ve sorunlarının çözümü; Soru= İçinde var denilen birçok şeyi göremiyorum bunun sebebi nedir? Cevap= B

Sexettin 16 Dec 17, 2022
DANet for Tabular data classification/ regression.

Deep Abstract Networks A PyTorch code implemented for the submission DANets: Deep Abstract Networks for Tabular Data Classification and Regression. Do

Ronnie Rocket 55 Sep 14, 2022
Fully convolutional networks for semantic segmentation

FCN-semantic-segmentation Simple end-to-end semantic segmentation using fully convolutional networks [1]. Takes a pretrained 34-layer ResNet [2], remo

Kai Arulkumaran 186 Dec 25, 2022
PyDeepFakeDet is an integrated and scalable tool for Deepfake detection.

PyDeepFakeDet An integrated and scalable library for Deepfake detection research. Introduction PyDeepFakeDet is an integrated and scalable Deepfake de

Junke, Wang 49 Dec 11, 2022
[Official] Exploring Temporal Coherence for More General Video Face Forgery Detection(ICCV 2021)

Exploring Temporal Coherence for More General Video Face Forgery Detection(FTCN) Yinglin Zheng, Jianmin Bao, Dong Chen, Ming Zeng, Fang Wen Accepted b

57 Dec 28, 2022
Code for ACL 2019 Paper: "COMET: Commonsense Transformers for Automatic Knowledge Graph Construction"

To run a generation experiment (either conceptnet or atomic), follow these instructions: First Steps First clone, the repo: git clone https://github.c

Antoine Bosselut 575 Jan 01, 2023
Paddle-Adversarial-Toolbox (PAT) is a Python library for Deep Learning Security based on PaddlePaddle.

Paddle-Adversarial-Toolbox Paddle-Adversarial-Toolbox (PAT) is a Python library for Deep Learning Security based on PaddlePaddle. Model Zoo Common FGS

AgentMaker 17 Nov 08, 2022
This package proposes simplified exporting pytorch models to ONNX and TensorRT, and also gives some base interface for model inference.

PyTorch Infer Utils This package proposes simplified exporting pytorch models to ONNX and TensorRT, and also gives some base interface for model infer

Alex Gorodnitskiy 11 Mar 20, 2022