Giter VIP home page Giter VIP logo

deltapointers's Introduction

About

This repository contains the source code for the EuroSys'18 paper "Delta Pointers: Buffer Overflow Checks Without the Checks" by Taddeus Kroes, Koen Koning (shared first authors), Erik van der Kouwe, Herbert Bos and Cristiano Giuffrida. The paper is available for download here.

Delta Pointers protect programs against buffer overflows by means of instrumentation inserted by the compiler. The inserted instrumentation inserts and updates a so-called delta tag into each pointer, which records two pieces of information: the distance of the pointer to the end of its corresponding memory object in the virtual address space, and a single overflow bit that indicates whether the pointer is out-of-bounds. The state of the overflow bit is managed efficiently by arithmetic operations, without the use of any additional branches or memory accesses. Upon dereference, the distance is taken out of the pointer using a bitwise AND operation, but the overflow bit is left intact. When set, this bit will cause the dereferenced pointer to become non-canonical, in turn causing a general protection fault in the processor. This way, out-of-bounds pointers are automatically rejected by the hardware.

Building and running instrumented programs

This repository only contains code for LLVM passes and a small runtime library. We use an external infrastructure library to plug these passes into existing build systems like that of SPEC.

First, make sure the infrastructure is up-to-date (in case you did not so a recursive clone of this repo):

$ git submodule update --init

The infrastructure's only hard dependency is Python 3.5. It downloads and builds most tools and libraries needed to build LLVM, but those in turn have some dependencies. On a clean Ubuntu 16.04 installation, this is what you need:

$ sudo apt-get install bison build-essential gettext git pkg-config python ssh

For nicer command-line usage, install the following python packages (optional):

$ pip3 install --user coloredlogs argcomplete

argcomplete enables command-line argument completion, but it needs to be activated first (optional):

$ eval "$(register-python-argcomplete --complete-arguments -o nospace -o default -- setup.py)"

Building/running benchmarks and dependencies is done with setup.py which knows about dependencies and build scripts. The following command builds all dependencies, the Delta Pointers passes and runtime, and some small test programs. It then runs the test programs, both without (the clang-lto baseline instance) and with our instrumentation (the deltatags instance):

$ ./setup.py run --build deltatags-test clang-lto deltatags

This will take a long time, so go get some coffee and plug in your laptop battery. The output should look somewhat like this. For clang-lto, the tests that do a buffer overflow should fail because they expect an error to be raised. For deltatags, all test should succeed.

To run SPEC-CPU2006, you will need to provide your own copy of the source. You can do so by modifying source and source_type on the relevant lines in setup.py. See the relevant documentation for details. After configuring setup.py, build and run the benchmark suite like this:

$ ./setup.py run --build spec2006 deltatags --test

This will build and run all 19 C/C++ benchmarks with Delta Pointers instrumentation, using the 'test' workload. To run the baseline, use clang-lto instead of deltatags. Results will be in the SPEC installation directory build/targets/spec2006/install. For a complete list of run options, consult:

$ ./setup.py run --help
$ ./setup.py run spec2006 --help

Instrumenting your own program

To run instrumentation on your own programs, you can either extract the relevant parts (llvm-passes/ and runtime/) and put them into your own repository, or you can define your own target. See target.py for an example.

Repo organization

The source consists of several components:

  • llvm-passes/ LLVM passes that instrument a program at compile time. This is the core of our work.

  • runtime/ A runtime library containing helper functions called by our instrumentation. dep.py informs the setup script how to build the runtime.

  • patches/ Some patches for the SPEC-CPU2006 benchmark suite, making it compatible with tagged pointers.

  • deltatags-test A collection of toy programs that test different parts of the Delta Pointers implementation. config/targets.py informs the setup script how to build/run these programs.

  • infra An external repository that facilitates program instrumentation for common benchmarks in systems research (developed in conjunction with Delta Pointers). The interface into this framework is setup.py.

  • setup.py is the main tool to build/run stuff with. This is where you register new benchmarks (or 'targets') and hook in any custom passes of your own. The script has descriptive usage messages for all subcommands.

  • config/instances.py informs the setup script how to build programs with Delta Pointers instrumentation. If you want to add custom instrumentation passes, this is the place to do it.

deltapointers's People

Contributors

taddeus avatar koenk avatar

Stargazers

Jin avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.