Note: The SymCC project is currently understaffed and therefore maintained in a best effort mode. In fact, we are hiring, in case you are interested to join the S3 group at Eurecom to work on this (and other projects in the group) please contact us. We nevertheless appreciate PRs and apologize in advance for the slow processing of PRs, we will try to merge them when possible.

SymCC: efficient compiler-based symbolic execution

SymCC is a compiler wrapper which embeds symbolic execution into the program during compilation, and an associated run-time support library. In essence, the compiler inserts code that computes symbolic expressions for each value in the program. The actual computation happens through calls to the support library at run time.

To build the pass and the support library, install LLVM (any version between 8 and 16) and Z3 (version 4.5 or later), as well as a C++ compiler with support for C++17. LLVM lit is only needed to run the tests; if it's not packaged with your LLVM, you can get it with pip install lit.

Under Ubuntu Groovy the following one liner should install all required packages:

sudo apt install -y git cargo clang-10 cmake g++ git libz3-dev llvm-10-dev llvm-10-tools ninja-build python2 python3-pip zlib1g-dev && sudo pip3 install lit

Alternatively, see below for using the provided Dockerfile, or the file util/quicktest.sh for exact steps to perform under Ubuntu (or use with the provided Vagrant file).

Make sure to pull the QSYM code:

$ git submodule init
$ git submodule update

Note that it is not necessary or recommended to build the QSYM submodule - our build system will automatically extract the right source files and include them in the build.

Create a build directory somewhere, and execute the following commands inside it:

$ cmake -G Ninja -DQSYM_BACKEND=ON /path/to/compiler/sources
$ ninja check

If LLVM is installed in a non-standard location, add the CMake parameter -DLLVM_DIR=/path/to/llvm/cmake/module. Similarly, you can point to a non-standard Z3 installation with -DZ3_DIR=/path/to/z3/cmake/module (which requires Z3 to be built with CMake).

The main build artifact from the user's point of view is symcc, a wrapper script around clang that sets the right options to load our pass and link against the run-time library. (See below for additional C++ support.)

To try the compiler, take some simple C code like the following:

#include <stdio.h>
#include <stdint.h>
#include <unistd.h>

int foo(int a, int b) {
    if (2 * a < b)
        return a;
    else if (a % b)
        return b;
    else
        return a + b;
}

int main(int argc, char* argv[]) {
    int x;
    if (read(STDIN_FILENO, &x, sizeof(x)) != sizeof(x)) {
        printf("Failed to read x\n");
        return -1;
    }
    printf("%d\n", foo(x, 7));
    return 0;
}

Save the code as test.c. To compile it with symbolic execution built in, we call symcc as we would normally call clang:

$ ./symcc test.c -o test

Before starting the analysis, create a directory for the results and tell SymCC about it:

$ mkdir results
$ export SYMCC_OUTPUT_DIR=`pwd`/results

Then run the program like any other binary, providing arbitrary input:

$ echo 'aaaa' | ./test

The program will execute the same computations as an uninstrumented version would, but additionally the injected code will track computations symbolically and attempt to compute diverging inputs at each branch point. All data that the program reads from standard input is treated as symbolic; alternatively, you can set the environment variable SYMCC_INPUT_FILE to the name of a file whose contents will be treated as symbolic when read.

Note that due to how the QSYM backend is implemented, all input has to be available from the start. In particular, when providing symbolic data on standard input interactively, you need to terminate your input by pressing Ctrl+D before the program starts to execute.

When execution is finished, the result directory will contain the new test cases generated during program execution. Try running the program again on one of those (or use util/pure_concolic_execution.sh to automate the process). For better results, combine SymCC with a fuzzer (see docs/Fuzzing.txt).

Documentation

The directory docs contains documentation on several internal aspects of SymCC, as well as building C++ code, compiling 32-bit binaries on a 64-bit host, and running SymCC with a fuzzer. There is also a list of all configuration options.

If you're interested in the research paper that we wrote about SymCC, have a look at our group's website. It also contains detailed instructions to replicate our experiments, as well as the raw results that we obtained.

Video demonstration

On YouTube you can find a practical introduction to SymCC as well as a video on how to combine AFL and SymCC

Building a Docker image

If you prefer a Docker container over building SymCC natively, just tell Docker to build the image after pulling the QSYM code as above. (Be warned though: the Docker image enables optional C++ support from source, so creating the image can take quite some time!)

$ git submodule init
$ git submodule update
$ docker build -t symcc .
$ docker run -it --rm symcc

This will build a Docker image and run an ephemeral container to try out SymCC. Inside the container, symcc is available as a drop-in replacement for clang, using the QSYM backend; similarly, sym++ can be used instead of clang++. Now try something like the following inside the container:

container$ cat sample.cpp
(Note that "root" is the input we're looking for.)
container$ sym++ -o sample sample.cpp
container$ echo test | ./sample
...
container$ cat /tmp/output/000008-optimistic
root

The Docker image also has AFL and symcc_fuzzing_helper preinstalled, so you can use it to run SymCC with a fuzzer as described in the docs. (The AFL binaries are located in /afl.)

While the Docker image is very convenient for using SymCC, I recommend a local build outside Docker for development. Docker will rebuild most of the image on every change to SymCC (which is, in principle the right thing to do), whereas in many cases it is sufficient to let the build system figure out what to rebuild (and recompile, e.g., libc++ only when necessary).

FAQ / BUGS / TODOs

Why is SymCC only exploring one path and not all paths?

SymCC is currently a concolic executor it follows the concrete path. In theory, it would be possible to make it a forking executor see issue #14

Why does SymCC not generate some test cases?

There are multiple possible reasons:

QSym backend performs pruning

When built with the QSym backend exploration (e.g., loops) symcc is subject to path pruning, this is part of the optimizations that makes SymCC/QSym fast, it isn't sound. This is not a problem for using in hybrid fuzzing, but this may be a problem for other uses. See for example issue #88.

When building with the simple backend the paths should be found. If the paths are not found with the simple backend this may be a bug (or possibly a limitation of the simple backend).

Incomplete symbolic handing of functions, systems interactions.

The current symbolic understanding of libc is incomplete. So when an unsupported libc function is called SymCC can't trace the computations that happen in the function.

1. Adding the function to the collection of wrapped libc functions and register the wrapper in the compiler.
2. Build a fully instrumented libc.
3. Cherry-pick individual libc functions from a libc implementation (e.g., musl)

See issue #23 for more details.

Rust support ?

This would be possible to support RUST, see issue #1 for tracking this.

Bug reporting

We appreciate bugs with test cases and steps to reproduce, PR with corresponding test cases. SymCC is currently understaffed, we hope to catch up and get back to active development at some point.

Contact

Feel free to use GitHub issues and pull requests for improvements, bug reports, etc. Alternatively, you can send an email to Sebastian Poeplau ([email protected]) and Aurélien Francillon ([email protected]).

Reference

To cite SymCC in scientific work, please use the following BibTeX:

@inproceedings {poeplau2020symcc,
  author =       {Sebastian Poeplau and Aurélien Francillon},
  title =        {Symbolic execution with {SymCC}: Don't interpret, compile!},
  booktitle =    {29th {USENIX} Security Symposium ({USENIX} Security 20)},
  isbn =         {978-1-939133-17-5},
  pages =        {181--198},
  year =         2020,
  url =          {https://www.usenix.org/conference/usenixsecurity20/presentation/poeplau},
  publisher =    {{USENIX} Association},
  month =        aug,
}

More information on the paper is available here.

License

SymCC is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

As an exception from the above, you can redistribute and/or modify the SymCC runtime under the terms of the GNU Lesser General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. See #114 for the rationale.

SymCC is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License and the GNU Lesser General Public License along with SymCC. If not, see https://www.gnu.org/licenses/.

The following pieces of software have additional or alternate copyrights, licenses, and/or restrictions: