Graphene-SGX Library OS: SGX supported guest OS that hosts CovaClave
Branch: master
Clone or download
Pull request Compare This branch is 116 commits behind oscarlab:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Graphene Library OS with Intel®️ SGX Support

Travis Logo

A Linux-compatible Library OS for Multi-Process Applications


Graphene Library OS is a project which provides lightweight guest OSes with support for Linux multi-process applications. Graphene can run applications in an isolated environment with virtualization benefits such as guest customization, platform independence, and migration, which is comparable to other virtual machines.

Graphene Library OS supports native, unmodified Linux applications on any platform. Currently, Graphene Library OS is successfully ported to Linux, FreeBSD and Intel SGX enclaves upon Linux platforms.

With the Intel®️ SGX support, Graphene Library OS can secure a critical application in a hardware encrypted memory region. Graphene Library OS can protect applications against a malicious system stack with minimal porting effort.

Graphene Library OS is a work published in Eurosys 2014. For more information. see the paper: Tsai, et al, "Cooperation and Security Isolation of Library OSes for Multi-Process Applications", Eurosys 2014.


Graphene Library OS is consist of five parts:

  • Instrumented GNU Library C
  • LibOS (a shared library named "")
  • PAL, a.k.a Platform Adaption Layer (a shared library named "")
  • Reference monitor (a shared library named "")
  • Minor kernel customization and kernel modules

Graphene Library OS currently only works on x86_64 architecture.

Graphene Library OS is tested to be compiling and running on Ubuntu 14.04/16.04 (both server and desktop version), along with Linux kernel 3.5/3.14/4.4. We recommand to build and install Graphene with the same host platform. Other distributions of 64-bit Linux can potentially, but the result is not guaranteed. If you find Graphene not working on other distributions, please contact us with a detailed bug report.

The following packages are required for building Graphene: (can be installed with 'apt-get install')

  • build-essential
  • autoconf
  • gawk
  • gcc 4 or 5

The following packages are also required for building Graphene for SGX (can be installed with 'apt-get install'):

  • python-protobuf
  • python-crypto

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

git submodule update --init

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

To build Graphene library OS with debug symbols, run "make DEBUG=1" instead of "make".


** Note: this step is optional. **

** Note: for building with Intel®️ SGX support, skip this step. **

** Disclaimer: this feature is experimental and may contain bugs. Please do no use in production system before further assessment.

To enable sandboxing, a customized Linux kernel is needed. Note that this feature is optional and completely unnecessary for running on SGX. To build the Graphene Linux kernel, do the following steps:

cd Pal/linux-3.19
make menuconfig
make install
(Add Graphene kernel as a boot option by commands like "update-grub")
(reboot and choose the Graphene kernel)

Please note that the building process may pause before building the Linux kernel, because it requires you to provide a sensible configuration file (.config). The Graphene kernel requires the following options to be enabled in the configuration:


For more details about the building and installation, see the Graphene github Wiki page:


To build Graphene Library OS with Intel SGX support, run "make SGX=1" instead of "make". "DEBUG=1" can be used to build with debug symbols. Using "make SGX=1" in the test or regression directory will automatically generate the enclave signatures (in .sig files).

A 3072-bit RSA private key (PEM format) is required for signing the enclaves. The default enclave key is placed in 'host/Linux-SGX/signer/enclave-key.pem', or the key can be specified through environment variable 'SGX_ENCLAVE_KEY' when building Graphene with Intel SGX support. If you don't have a private key, create it with the following command:

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

After signing the enclaves, users may ship the application files with the built Graphene Library OS, along with a SGX-specific manifest (.manifest.sgx files) and the signatures, to the Intel SGX-enabled hosts. The Intel SGX Linux SDK is required for running Graphene Library OS. Download and install from the official Intel github repositories:

(The SDK and driver version must be 1.9 or LOWER)

A Linux driver must be installed before runing Graphene Library OS in enclaves. Simply run the following command to build the driver:

** Please make sure the GCC version is either 4 or 5 **

cd Pal/src/host/Linux-SGX/sgx-driver
(The console will be prompted to ask for the path of Intel SGX driver code)
sudo ./

Finally generating the runtime enclave tokens by running "make SGX_RUN=1".


Graphene library OS uses PAL ( as a loader to bootstrap an application in the library OS. To start Graphene, PAL ( will have to be run as an executable, with the name of the program, and a "manifest file" 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)
(Manifest must have "#![PATH_TO_PAL]/" as the first line)

Using "" as loader to start Graphene will not attach the applications to the Graphene reference monitor. Tha applications will have better performance, but no strong security isolation. To attach the applications to the Graphene reference monitor, Graphene must be started with the PAL reference monitor loader ( Graphene provides three options for specifying the programs and manifest files to the loader:

  • option 4: (automatic manifest - with reference monitor)
SEC=1 [PATH TO Runtime]/pal_loader [PROGRAM] [ARGUMENTS]...
(Manifest file: "[PROGRAM].manifest" or "manifest")
  • option 5: (given manifest - with reference monitor)
SEC=1 [PATH TO Pal/src]/pal_loader [MANIFEST] [ARGUMENTS]...
  • option 6: (manifest as a script - with reference monitor)
(Manifest must have "#![PATH TO Pal/src]/pal_sec" as the first line)

Although manifest files are optional for Graphene, running an application usually requires some minimal configuration in its manifest file. A sensible manifest file will include paths to the library OS and GNU library C, environment variables such as LD_LIBRARY_PATH, file systems to be mounted, and isolation rules to be enforced in the reference monitor.

Here is an example of manifest files:

loader.preload = file:LibOS/shim/src/
loader.env.LDL_LIBRAY_PATH = /lib
fs.mount.glibc.type = chroot
fs.mount.glibc.path = /lib
fs.mount.glibc.uri = file:LibOS/build

More examples can be found in the test directories (LibOS/shim/test). We have also tested several commercial applications such as GCC, Bash and Apache, and the manifest files that bootstrap them in Graphene are provided in the individual directories.

For more information and the detail of the manifest syntax, see the Graphene github Wiki page:


For any questions or bug reports, please contact us:

or post an issue on our github repository: