PANDA is an open-source Platform for Architecture-Neutral Dynamic Analysis. It is built upon the QEMU whole system emulator, and so analyses have access to all code executing in the guest and all data. PANDA adds the ability to record and replay executions, enabling iterative, deep, whole system analyses. Further, the replay log files are compact and shareable, allowing for repeatable experiments. A nine billion instruction boot of FreeBSD, e.g., is represented by only a few hundred MB. PANDA leverages QEMU's support of thirteen different CPU architectures to make analyses of those diverse instruction sets possible within the LLVM IR. In this way, PANDA can have a single dynamic taint analysis, for example, that precisely supports many CPUs. PANDA analyses are written in a simple plugin architecture which includes a mechanism to share functionality between plugins, increasing analysis code re-use and simplifying complex analysis development.
It is currently being developed in collaboration with MIT Lincoln Laboratory, NYU, and Northeastern University. PANDA is released under the GPLv2 license.
We have two primary branches of PANDA:
dev for development and
stable for stable+versioned releases. To learn more about the differences between these branches and version numbers, visit our wiki. In general, PANDA resources (i.e., docker containers and documentation) are based off the
dev branch. We recommend using the
stable branch if you're going to fork the project and later pull in updates.
The latest version of PANDA's
master branch is automatically built as a two docker images based on Ubuntu 20.04 and published to Docker Hub.
Most users will want to use the
panda container which has PANDA and PyPANDA installed along with their runtime dependencies, but no build artifacts or source code to reduce the size of the container.
Developers interested in using Docker should use the
pandadev container which has PANDA and PyPANDA installed, build and runtime dependencies for both, all build artifacts and source code and the contents of this repository in the
To use the
panda container you can pull it from Docker Hub:
$ docker pull pandare/panda $ docker run --rm pandare/panda panda-system-i386 --help
Or build from this repository:
$ DOCKER_BUILDKIT=1 docker build --target=panda -t panda . $ docker run --rm panda panda-system-i386 --help
To use the
pandadev container, you can pull it from Docker Hub:
$ docker pull pandare/pandadev $ docker run --rm pandare/pandadev /panda/build/panda-system-i386 --help
Or build from this repository:
$ DOCKER_BUILDKIT=1 docker build --target=developer -t pandadev . $ docker run --rm pandadev panda-system-i386 --help
Quickstart: Python pip
The Python interface to PANDA (also known as pypanda) can be installed from PIP by running
pip3 install pandare. This will install everything you need for python-based PANDA analyses, but not stand-alone PANDA binaries. This package is not automatically updated so it may fall behind the master branch of PANDA. The distributed binaries are only tested on 64-bit Ubuntu 18.04 and other architectures/versions are unlikely to work. You can also install pypanda by building PANDA and then running
python3 setup.py install from the directory
Because PANDA has a few dependencies, we've encoded the build instructions into the install_ubuntu.sh. The script should work on the latest Debian stable/Ubuntu LTS versions. If you wish to build PANDA manually, you can also check the step-by-step instructions in the documentation directory.
We currently only vouch for buildability on the latest Debian stable/Ubuntu LTS, but we welcome pull requests to fix issues with other distros.
For other distributions, it should be straightforward to translate the
commands into whatever package manager your distribution uses.
Note that if you want to use our LLVM features (mainly the dynamic taint
system), you will need to install LLVM 11 from OS packages or compiled from
source. On Ubuntu this should happen automatically via
Additionally, it is strongly recommended that you only build PANDA as 64bit
binary. Creating a 32bit build should be possible, but best avoided.
See the limitations section for details.
The install_arch.sh has been contributed for building PANDA on Arch Linux. Currently, the script has only been tested on Arch Linux 4.17.5-1-MANJARO. You can also find step-by-step instructions for building on Arch in the documentation directory.
Building on Mac is less well-tested, but has been known to work. There is a script, install_osx.sh to build under OS X. The script uses homebrew to install the PANDA dependencies. As homebrew is known to be very fast in deprecating support for older versions of OS X and supported packages, expect this to be broken.
After successfully building PANDA, you can copy the build to a system-wide
location by running
make install. The default installation path is
You can specify an alternate installation path through the
--prefix=/opt/panda. Note that your system must have
installed in order for
make install to succeed.
bin directory containing the PANDA binaries is in your
variable, then you can run PANDA similarly to QEMU:
panda-system-i386 -m 2G -hda guest.img -monitor stdio
PANDA uses the LLVM architecture from the S2E project.
This allows translating the TCG intermediate code representation used by QEMU,
to LLVM IR. The latter has the advantages of being easier to work with, as well
as platform independent. This enables the implementation of complex analyses
The S2E files used by PANDA to support taint analysis have been updated to work with LLVM 11.
Great effort is put to maintain the PANDA trace format stable so that existing
traces remain replayable in the future. Changes that will break existing traces
However, currently, record/replay is only guaranteed between PANDA builds of the
same address length. E.g. you can't replay a trace captured on a 32bit build of
PANDA on a 64bit of PANDA. The reason for this is that some raw pointers managed
to creep into the trace format (see headers in
Given the memory limitations of 32bit builds, almost all PANDA users use 64bit. As a result, this issue should affect only a tiny minority of users. This is also supported by the fact that the issue remained unreported for a long time (>3 years). Therefore, when a fix is to be implemented, it may be assessed that migrating existing recordings captured by 32bit builds is not worth the effort.
For this, it is strongly recommended that you only create and use 64bit builds of PANDA. If you happen to already have a dataset of traces captured by a 32bit build of PANDA, you should contact the community ASAP to discuss possible options.
Documentation and Support
PANDA currently supports whole-system record/replay execution, as well as time-travel debugging, of x86, x86_64, and ARM guests. Other architectures (mips, mipsel, ppc) may be run under PANDA without record/replay support. Details about the implementation and use of PANDA can be found in the PANDA manual. Some of the topics covered are:
- details about record/replay
- the architecture-neutral plugin interface
- the callbacks provided by PANDA
- plugin zoo
- python interface
Documentation for individual plugins is provided by the
in the plugin directory. See panda/plugins directory.
 B. Dolan-Gavitt, T. Leek, J. Hodosh, W. Lee. Tappan Zee (North) Bridge: Mining Memory Accesses for Introspection. 20th ACM Conference on Computer and Communications Security (CCS), Berlin, Germany, November 2013.
 R. Whelan, T. Leek, D. Kaeli. Architecture-Independent Dynamic Information Flow Tracking. 22nd International Conference on Compiler Construction (CC), Rome, Italy, March 2013.
 B. Dolan-Gavitt, J. Hodosh, P. Hulin, T. Leek, R. Whelan. Repeatable Reverse Engineering with PANDA. 5th Program Protection and Reverse Engineering Workshop, Los Angeles, California, December 2015.
 M. Stamatogiannakis, P. Groth, H. Bos. Decoupling Provenance Capture and Analysis from Execution. 7th USENIX Workshop on the Theory and Practice of Provenance, Edinburgh, Scotland, July 2015.
 B. Dolan-Gavitt, P. Hulin, T. Leek, E. Kirda, A. Mambretti, W. Robertson, F. Ulrich, R. Whelan. LAVA: Large-scale Automated Vulnerability Addition. 37th IEEE Symposium on Security and Privacy, San Jose, California, May 2016.
This material is based upon work supported under Air Force Contract No. FA8721-05-C-0002 and/or FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the U.S. Air Force.