/grr

High-throughput fuzzer and emulator of DECREE binaries

Primary LanguageC++Apache License 2.0Apache-2.0

GRR

Build Status Slack Chat

GRR is pronounced with two fists in the air

GRR is an x86 to amd64 binary translator. GRR was created to emulate and fuzz DECREE challenge binaries.

Features

  • Code cache persistence (avoids translation overheads across separate runs).
  • Optimization of the code cache, including the persisted cache.
  • Multi-processing support (allows multiple communicating, 32-bit processes to be emulated within a single 64-bit address space).
  • Built-in fuzzing engine, which includes Radamsa.
  • Support for self-modifying code (e.g. JIT compilers). Self-modified code can also be persisted.

Anti-features

  • Orchestration. GRR does not manage a fuzzing campaign. An external orchestration system can direct GRR to fuzz a program, and identify a minimal set of maximal coverage-producing inputs.

Compiling

Install System dependcies

sudo apt-get install -y git build-essential
sudo apt-get install -y clang llvm-dev libc++1 libc++-dev

Then we can build GRR. This script will create grr-build in the current working directory, and download remaining dependencies.

./grr/scripts/build.sh

Running

There are two steps to running GRR: snapshotting, and record/replaying. Snapshotting creates an initial image of the binary after execve. Record/replaying takes an input testcase and attempts to replay that testcase. The replay can be recorded or mutated as well.

Snapshotting

You can run grrshot on a CGC challenge binary, or on a statically linked 32-bit ELF version of a CGC challenge binary.

./bin/debug_linux_user/grrshot --num_exe=1 --exe_dir=/path --exe_prefix=CADET_000 --snapshot_dir=/tmp/snapshot

This will create a snapshot of /path/CADET_00001 and store the snapshot into the /tmp/snapshot directory.

Replaying

./bin/debug_linux_user/grrplay --num_exe=1 --snapshot_dir=/tmp/snapshot --persist_dir=/tmp/persist --input=/path/to/testcase 

Replay + Recording

./bin/debug_linux_user/grrplay --num_exe=1 --snapshot_dir=/tmp/snapshot --persist_dir=/tmp/persist --input=/path/to/testcase --output_dir=/tmp/out 

Replay + Recording + Mutating

./bin/debug_linux_user/grrplay --num_exe=1 --snapshot_dir=/tmp/snapshot --persist_dir=/tmp/persist --input=/path/to/testcase --output_dir=/tmp/out --input_mutator=inf_radamsa_spliced

There are many mutators. Some of the mutators are deterministic, and therefore run for a period of time that is proportional to the number of receive system calls in the input testcase. Other mutators are non-deterministic and can run forever. These mutators are prefixed with inf_.

Dependencies

Intel XED

GRR depends on the Intel XED instruction encoder/decoder. XED is licensed under the Apache License.

Radamsa

GRR depends on Radamsa, a high-quality input mutation engine. GRR embeds a version of Radamsa that can be used as a kind of library. Radamsa is licensed under the MIT license.

Other dependencies

GRR depends on gflags.

GRR depends on Frank Thilo's C++ port of the the RSA Data Security, Inc. MD5 Message-Digest Algorithm.

GRR uses Yann Collet's xxHash as part of its code cache index hash table.

History

GRR, short for Granary Record/Replay, is the third iteration of the Granary series of dynamic binary translators. GRR is pgoodman's third dynamic binary translator.

GRR was created for the DARPA Cyber Grand Challenge. It was an essential component of Trail of Bit's CRS. The version of GRR in this repository differs from the CRS version in some important ways.

Don't tread on me!