Graphene Library OS with Intel SGX Support

A Linux-compatible Library OS for Multi-Process Applications

What is Graphene?

Graphene is a lightweight guest OS, designed to run a single application with minimal host requirements. Graphene can run applications in an isolated environment with benefits comparable to running a complete OS in a virtual machine -- including guest customization, ease of porting to different OSes, and process migration.

Graphene supports native, unmodified Linux applications on any platform. Currently, Graphene runs on Linux and Intel SGX enclaves on Linux platforms.

With Intel SGX support, Graphene can secure a critical application in a hardware-encrypted memory region. Graphene can protect applications from a malicious system stack with minimal porting effort.

Our papers describe the motivation, design choices, and measured performance of Graphene:

• EuroSys 2014
• ATC 2017

How to build Graphene?

Graphene consists of three parts:

• An instrumented GNU C Library
• The Library OS itself (a shared library named libsysdb.so, called the "shim" in our source code)
• The Platform Adaptation Layer, or PAL, (a shared library named libpal.so)

Graphene currently only works on the x86_64 architecture. Graphene is currently tested on Ubuntu 16.04 and 18.04 (both server and desktop version), along with Linux kernel versions 3.x/4.x. We recommend building and installing Graphene on the same host platform. If you find problems with Graphene on other Linux distributions, please contact us with a detailed bug report.

Run the following command on Ubuntu to install dependencies for Graphene:

sudo apt-get install -y build-essential autoconf gawk bison

For building Graphene for SGX, run the following command in addition:

sudo apt-get install -y python3-protobuf libprotobuf-c-dev protobuf-c-compiler

To run tests locally, you also need the python3-pytest package:

sudo apt-get install -y python3-pytest

To build Graphene, simply run the following commands in the root of the source tree:

git submodule update --init -- Pal/src/host/Linux-SGX/sgx-driver/
make

Each part of Graphene can be built separately in the subdirectories.

To build Graphene with debug symbols, run make DEBUG=1 instead of make. To specify custom mirrors for downloading the GLIBC source, use make GLIBC_MIRRORS=....

To build with -Werror, run make WERROR=1.

Building with kernel-level sandboxing (optional)

This feature is marked as EXPERIMENTAL and no longer exists on the master branch.

Building with Intel SGX Support

Prerequisites

1. Generate signing keys

   A 3072-bit RSA private key (PEM format) is required for signing the manifest. If you don't have a private key, create it with the following command:

   openssl genrsa -3 -out enclave-key.pem 3072
   

   You can either place the generated enclave key in the default path, host/Linux-SGX/signer/enclave-key.pem, or specify the key's location through the environment variable SGX_SIGNER_KEY.

   After signing the application's manifest, users may ship the application and Graphene binaries, along with an SGX-specific manifest (.manifest.sgx extension), the signature (.sig extension), and the aesmd init token (.token extension) to execute on another SGX-enabled host.

2. Install the Intel SGX SDK and driver

   The Intel SGX Linux SDK is required to compile and run Graphene on SGX. Download and install it from the official Intel GitHub repositories:

   • <https://github.com/01org/linux-sgx>
   • <https://github.com/01org/linux-sgx-driver>

3. Build and install the Graphene SGX driver A Graphene-specific Linux driver must also be installed before running Graphene in an SGX environment. Simply run the following commands to build the driver:

   cd Pal/src/host/Linux-SGX/sgx-driver
   make
   # The console will be prompted to ask for the path of Intel SGX driver code
   sudo ./load.sh
   sudo sysctl vm.mmap_min_addr = 0
   

   We note that this last command is a tempoarary work-around for some issues with the Intel SGX driver. This is an inadvisable configuration for production systems. We hope to remove this step in a future version of Graphene, once the SGX driver is upstreamed to Linux.

Building Graphene-SGX

To build Graphene with Intel SGX support, in the root directory of Graphene repo, run the following command:

make SGX=1

To build with debug symbols, instead run the command:

make SGX=1 DEBUG=1

Running make SGX=1 in the test or regression directory will automatically generate the required manifest signatures (.sig files).

Run Built-in Examples in Graphene-SGX

There are a few built-in examples under LibOS/shim/test/. The "native" folder includes a rich set of C programs and "apps" folder includes a few tested applications, such as GCC, Python, and Apache.

1. Build and run a helloworld program with Graphene-SGX

   • go to LibOS/shim/test/native, build the enclaves via the command:

     make SGX=1
     

     This command will build enclaves for all the programs in the folder

   • Generate the token from aesmd service, via the command:

     make SGX=1 sgx-tokens
     

   • Run a helloworld program with Graphene-SGX:

     SGX=1 ./pal_loader helloworld
     

     or:

     ./pal_loader SGX helloworld
     

2. Build and run the Python helloworld.py script in Graphene-SGX

   • go to LibOS/shim/test/apps/python, and build the enclave:

     make SGX=1
     

   • Generate a launch token from the aesmd service:

     make SGX=1 sgx-tokens
     

   • Run helloworld.py script with Graphene-SGX via:

     SGX=1 ./python.manifest.sgx scripts/helloworld.py
     

Including Application Test Cases

To add the application test cases, issue the following command from the root of the source tree:

git submodule update --init -- LibOS/shim/test/apps/

Testing the remote attestation feature

To enable tests for the built-in remote attestation feature for Graphene-SGX, obtain a SPID and a subscription key (can be linkable or unlinkable) from the Intel API Portal: <https://api.portal.trustedservices.intel.com/EPID-attestation>.

Specify the SPID, subscription key, and the type of the SPID/key in the manifest:

sgx.ra_client_spid = <SPID>
sgx.ra_client_key = <KEY>
sgx.ra_client_linkable = 1 # or 0 if the SPID/key is unlinkable (default)

If the remote attestation feature is enabled, Graphene-SGX will terminate if the platform is not successfully verified by the Intel Attestation Service (IAS). The feature ensures that Graphene-SGX only executes on genuine, up-to-date SGX hardware.

To enable remote attestation tests in Pal/regression, specify the following variables:

cd PAL/regression
make SGX=1 RA_CLIENT_SPID=<SPID> RA_CLIENT_KEY=<KEY>
make SGX=1 sgx-tokens

If you receive a GROUP_OUT_OF_DATE status from IAS, this status indicates that your CPU is out of date and can be vulnerable to hardware attacks. If you wish to bypass this error, you can specify the following option in the manifest:

sgx.ra_accept_group_out_of_date = 1

Similarly, if you receive a CONFIGURATION_NEEDED status from IAS, this status indicates that additional configuration of your SGX platform may be needed. If you wish to bypass this error, you can specify the following option in the manifest:

sgx.ra_accept_configuration_needed = 1

Security advisories:

• GROUP_OUT_OF_DATE may indicate that the firmware (microcode) of you CPU is not updated according to INTEL-SA-00233 (Load/store data sampling) and INTEL-SA-00161 (L1 terminal fault). It is recommended that you keep the BIOS of your platform up-to-date.
• If you receive status CONFIGURATION_NEEDED from the IAS after updating your BIOS, you may need to disable hyperthreading in your BIOS to mitigate L1 terminal fault.

How to run an application in Graphene?

Graphene library OS uses the PAL (libpal.so) as a loader to bootstrap applications in the library OS. To start Graphene, PAL (libpal.so) will have to be run as an executable, with the name of the program, and a "manifest file" (per-app configuration) given from the command line. Graphene provides three options for specifying the programs and manifest files:

• option 1 (automatic manifest):

  [PATH TO Runtime]/pal_loader [PROGRAM] [ARGUMENTS]...
  (Manifest file: "[PROGRAM].manifest" or "manifest")
  

• option 2 (given manifest):

  [PATH TO Runtime]/pal_loader [MANIFEST] [ARGUMENTS]...
  

• option 3 (manifest as a script):

  [PATH TO MANIFEST]/[MANIFEST] [ARGUMENTS]...
  (Manifest must have "#![PATH_TO_PAL]/libpal.so" as the first line)
  

Running an application requires some minimal configuration in the application's manifest file. A sensible manifest file will include paths to the library OS and other libraries the application requires; environment variables, such as LD_LIBRARY_PATH; and file systems to be mounted.

Here is an example manifest file:

loader.preload = file:LibOS/shim/src/libsysdb.so
loader.env.LD_LIBRAY_PATH = /lib
fs.mount.libc.type = chroot
fs.mount.libc.path = /lib
fs.mount.libc.uri = file:[relative path to Graphene root]/Runtime

More examples can be found in the test directories (LibOS/shim/test). We have also tested several applications, such as GCC, Bash, and Apache. The manifest files for these applications are provided in the individual directories under LibOS/shim/test/apps.

For the full documentation of the Graphene manifest syntax, see the Graphene documentation.

Docker support

We are actively working on adding a proper Docker support. You can find the old and deprecated implementation on DEPRECATED/gsc branch (caveat: use at your own risk!).

Contact

For any questions or bug reports, please send an email to <[email protected]> or post an issue on our GitHub repository: <https://github.com/oscarlab/graphene/issues>.

Our mailing list is publicly archived here.