libdft: Practical Dynamic Data Flow Tracking

These code is modified from VUzzer64, and it is originally from libdft.

Features

• Support Intel Pin 3.x
• Support Intel 64 bit platform
• Support basic SSE, AVX instructions.
• Use BDD data structure described in Angora's paper.

Limitation of our taint propagation rules

• Byte level
• Ignore implicit flows
• Ignore eflags registers

TODO

• ternary instructions
• performance optimization
• support more instructions
• test for each instruction
• rules for eflags registers
• FPU instructions

Contributing guidance

As TaintInduce mentioned, libdft exists the soundness and completeness probelm.

• The taint propagation rules may be wrong.
• It only supports the basic instructions, and there are many other instructions it need to support.

If you want to contribute to this, modify the instructions in src/libdft_core.cpp, and pull requests on github for us.

Build

• Download Intel Pin 3.x and set PIN_ROOT to Pin's directory.

PREFIX=/path-to-install ./install_pin.sh

• build libdft64

make

Docker

docker build -t libdft ./
docker run --privileged -v /path-to-dir:/data -it --rm libdft /bin/bash

Introduction

Dynamic data flow tracking (DFT) deals with the tagging and tracking of "interesting" data as they propagate during program execution. DFT has been repeatedly implemented by a variety of tools for numerous purposes, including protection from buffer overflow and cross-site scripting attacks, analysis of legitimate and malicious software, detection and prevention of information leaks, etc. libdft is a dynamic DFT framework that is at once fast, reusable, and works with commodity software and hardware. It provides an API, which can be used to deliver DFT-enabled tools that can be applied on unmodified binaries running on common operating systems and hardware, thus facilitating research and rapid prototyping.

Installation & Usage

libdft relies on Intel Pin, which is a dynamic binary instrumentation (DBI) framework from Intel. In order to install libdft you first need a working copy on the latest Pin build, as well as the essential build tools for GNU/Linux (i.e., GCC, GNU Make, etc). After downloading and installing Intel Pin please follow the libdft installation instructions.

Tools

libdft is designed to facilitate the creation of "Pin tools" that employ dynamic DFT. As the name implies, libdft is also a shared library, which can be used to transparently perform DFT on binaries. Additionally, it provides an API that enables tool authors to adjust the applied DFT by specifying data sources and sinks, and customize the tag propagation policy. We have included three simple Pin tools inside the tools subdirectory to aid the development of DFT-powered Pintools:

• nullpin is essentially a null tool that runs a process using Pin without any form of instrumentation or analysis. This tool can be used to measure the overhead imposed by Pin's runtime environment.
• libdft uses libdft to apply DFT on the application being executed, but does not use any of the API functions to define data sources and sinks (i.e., it does not customize the applied DFT). This tool can be used to evaluate the overhead imposed by libdft.
• libdft-dta is an example tool that uses the API of libdft, and serves as template for future meta-tools. In particular, it implements a dynamic taint analysis (DTA) platform by transparently utilizing DFT in unmodified x86 Linux binaries.

DTA operates by tagging all data coming from the network as "tainted", tracking their propagation, and alerting the user when they are used in a way that could compromise his system. In this case, the network is the source of "interesting" data, while instructions that are used to control a program's flow are the sinks. For the x86 architecture, these are jumps and function calls with non-immediate operands, as well as function returns. Oftentimes, attackers are able to manipulate the operands of such instructions by abusing various types of software memory errors such as buffer overflows, format string vulnerabilities, dangling pointers, etc. They can then seize control of a program by redirecting execution to existing code (e.g., return-to-libc, ROP), or their own injected instructions. libdft-dta checks if tainted data are used in indirect control transfers, and if so, it halts execution with an informative message containing the offending instruction and the contents of the instruction pointer EIP.

Usage

After building both libdft and the accompanying tools (i.e., nullpin, libdft, and libdft-dta), you can apply them directly in unmodified x86 Linux binaries as follows (assuming that you have added Pin's location to your PATH, and installed libdft in your home directory):

pin -follow_execv -t ~/libdft/tools/nullpin.so -- <executable>

Arguments processed by Pin

• -follow_execv: Instructs Pin to also instrument all processes spawned using the exec(3) class system calls by the program.
• -t: Specifies the Pin tool to be used.

Research

Following are some publications that rely on libdft:

• libdft: Practical Dynamic Data Flow Tracking for Commodity Systems. Vasileios P. Kemerlis, Georgios Portokalidis, Kangkook Jee, and Angelos D. Keromytis. In Proceedings of the 8th ACM SIGPLAN/SIGOPS International Conference on Virtual Execution Environments (VEE). March 2012, London, UK.

• A General Approach for Efficiently Accelerating Software-based Dynamic Data Flow Tracking on Commodity Hardware. Kangkook Jee, Georgios Portokalidis, Vasileios P. Kemerlis, Soumyadeep Ghosh, David I. August, and Angelos D. Keromytis. In Proceedings of the 19th Internet Society (ISOC) Symposium on Network and Distributed System Security (NDSS). February 2012, San Diego, CA.

• Angora: Efficient Fuzzing by Principled Search In Proceedings of the IEEE Symposium on Security and Privacy (SP). San Francisco, CA, May 2018.

• VUzzer: Application-aware Evolutionary Fuzzing In Proceedings of the Internet Society (ISOC) Symposium on Network and Distributed System Security (NDSS). Feb 2017.

---

libdft64 is being developed by ByteDance AI Lab currently.