Skip to content

Detecting Spectre vulnerabilities using symbolic execution, built on angr (github.com/angr/angr)

License

Notifications You must be signed in to change notification settings

cdisselkoen/pitchfork

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

What is Pitchfork?

Pitchfork is a static analysis tool, built on angr, which performs speculative symbolic execution. That is, it not only executes the "correct" or "sequential" paths of a program, but also the "mispredicted" or "speculative" paths, subject to some speculation window size. Pitchfork finds paths where secret data is used in either address calculations or branch conditions (and thus leaked), even speculatively - these paths represent Spectre vulnerabilities. Pitchfork covers Spectre v1, Spectre v1.1, and Spectre v4.

Installing

  • Clone this repo:
    git clone https://github.com/cdisselkoen/pitchfork
  • Install the pypy JITting Python interpreter (highly recommended), rather than using the standard Python interpreter. You'll have to set that up first, before continuing with installation. E.g., on Mac:
    brew install pypy3
    cd /path/to/where/you/cloned/pitchfork
    pypy3 -m venv [whatever_you_want_to_name_your_virtualenv]
    source [your_virtualenv_name]/bin/activate
    then proceed with the following steps using the pypy virtualenv. Inside this virtualenv, python should give you a pypy interpreter rather than an ordinary Python one.
    • Pitchfork has only been tested with PyPy 7.0-7.1 (both of which implement Python 3.6.1), but it should also work with other PyPy versions and even other Python interpreters.
  • Install angr using the directions here
    • Pitchfork currently only works with angr version 8.19.4.5 (latest in Pip as of this writing)
    • You won't actually be using this angr, as Pitchfork requires a slightly patched version of angr which we'll install next. But performing the typical angr install process ensures that all dependencies (Python and otherwise) for angr 8.19.4.5 are properly installed
  • Clone our fork of angr inside this pitchfork directory, checking out its more-hooks branch:
    git clone https://github.com/cdisselkoen/angr -b more-hooks angr-git
    ln -s angr-git/angr angr
    • We recommend cloning our angr fork inside this pitchfork directory so that Python automatically uses it instead of the pip-installed angr. If you clone our angr fork somewhere else, you'll have to put that location on your PYTHONPATH and do pip uninstall angr (which will uninstall angr itself but leave all of its other dependencies in place).

Running

  • Make sure you activate your Python virtualenv:
    cd pitchfork
    source [your_virtualenv_name]/bin/activate
  • Run python pitchfork.py to run tests against all the original Kocher test cases, our new Spectre v1 test cases, and our Spectre v1.1 test cases (see below).
  • Explanations of expected Kocher test results are in kocher_analysis.txt (see also the new Spectre testcases)
  • To run other tests or workloads, look at the functions in pitchfork.py or eval.py
  • Some useful utilities for interactive investigation are in interactiveutils.py (imported with pitchfork.py)

Our Spectre testcases

We have three sets of Spectre testcases:

  • The original well-known Kocher testcases for Spectre v1. We are using the versions from Spectector; both the sources and binaries can be found in the spectector-clang folder.

  • A revised version of the Kocher testcases. The main difference in our revision is when executed non-speculatively, our revised testcases do not perform out-of-bounds or secret-dependent memory accesses, or branch on secret data. These cases are found in new-testcases/spectrev1.c; detailed explanations of changes can be found in comments in that file.

  • A new set of testcases for Spectre v1.1. Spectre v1.1 is similar to Spectre v1 except that it relies on out-of-bounds writes rather than out-of-bounds reads. Our testcases demonstrate a variety of ways that these vulnerabilities can appear. They are found in new-testcases/forwarding.c; detailed explanations of each of the cases can be found in comments in that file.

About

Detecting Spectre vulnerabilities using symbolic execution, built on angr (github.com/angr/angr)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •