Building

Todo

This page really belongs to devel/, move it there after a proper release. Instead, for all users, there should be documentation for installing without full compilation.

Graphene consists of several components:

  • 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)
  • A patched GNU C Library (a set of shared libraries libc.so, libpthread.so, libm.so, etc.)

The build of Graphene implies building at least the first two components. The build of the patched C library is optional but highly recommended for performance reasons. The patched C library is built by default.

Graphene currently only works on the x86_64 architecture. Graphene is currently tested on Ubuntu 20.04 and 18.04, along with Linux kernel version 5.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.

Installing dependencies

Common dependencies

Run the following command on Ubuntu to install dependencies:

sudo apt-get install -y \
    build-essential \
    autoconf \
    bison \
    gawk \
    meson \
    python3-click \
    python3-jinja2

For GDB support and to run all tests locally you also need to install:

sudo apt-get install -y python3-pyelftools python3-pytest libunwind8

Dependencies for SGX

The build of Graphene with SGX support requires the corresponding SGX software infrastructure to be installed on the system. In particular, the FSGSBASE functionality must be enabled in the Linux kernel, the Intel SGX driver must be running, and Intel SGX SDK/PSW/DCAP must be installed. In the future, when all required SGX infrastructure is upstreamed in Linux and popular Linux distributions, the prerequisite steps will be significantly simplified.

1. Required packages

Run the following commands on Ubuntu to install SGX-related dependencies:

sudo apt-get install -y \
    libcurl4-openssl-dev \
    libprotobuf-c-dev \
    protobuf-c-compiler \
    python3-pip \
    python3-protobuf
python3 -m pip install toml>=0.10

2. Install the Linux kernel patched with FSGSBASE

FSGSBASE is a feature in recent processors which allows direct access to the FS and GS segment base addresses. For more information about FSGSBASE and its benefits, see this discussion. FSGSBASE patchset was merged in 5.9. For older kernels it is available as separate patches.

The following instructions to patch and compile a Linux kernel with FSGSBASE support below are written around Ubuntu 18.04 LTS (Bionic Beaver) with a Linux 5.4 LTS stable kernel but can be adapted for other distros as necessary. These instructions ensure that the resulting kernel has FSGSBASE support and up to date security mitigations.

  1. Clone the repository with patches:

    git clone https://github.com/oscarlab/graphene-sgx-driver
    
  2. Setup a build environment for kernel development following the instructions in the Ubuntu wiki. Clone Linux version 5.4 via:

    git clone --single-branch --branch linux-5.4.y \
        https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
    cd linux
    
  3. Apply the provided FSGSBASE patches to the kernel source tree:

    git am <graphene-sgx-driver>/fsgsbase_patches/*.patch
    

    The conversation regarding this patchset can be found in the kernel mailing list archives here.

  4. Build and install the kernel following the instructions in the Ubuntu wiki.

  5. After rebooting, verify the patched kernel is the one that has been booted and is running:

    uname -r
    
  6. Also verify that the patched kernel supports FSGSBASE (the below command must return that bit 2 is set):

    LD_SHOW_AUXV=1 /bin/true | grep AT_HWCAP2
    

After the patched Linux kernel is installed, you may proceed with installations of other SGX software infrastructure: the Intel SGX Linux driver, the Intel SGX SDK/PSW, and Graphene itself (see next steps). Note that older versions of these software packages may not work with recent Linux kernels like 5.4. We recommend to use commit b7ccf6f of the Intel SGX Linux Driver for Intel SGX DCAP and commit 0e71c22 of the Intel SGX SDK/PSW.

3. Install the Intel SGX driver and SDK/PSW

This step depends on your hardware and kernel version.

If your CPU supports FLC, we recommend you install kernel 5.11 or later. The SGX driver is part of the upstream kernel. You only need to install SDK/PSW from one of the following choices.

If you have an older CPU without FLC support, you need to install the Intel SGX Linux SDK and the Intel SGX driver. Download and install them from the official Intel GitHub repositories:

Alternatively, if your CPU supports FLC, you can choose to install DCAP versions of the SDK and driver, download and install them from:

4. 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, Pal/src/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.

Building

Note

We’re in the middle of the migration from Make to Meson. In the meantime you need to run both buildchains, first make then meson.

To build Graphene, in the root directory of Graphene repo, run the following commands:

# if you build graphene-direct (note that "direct" means non-SGX version)
make

# if you build graphene-sgx
make SGX=1 ISGX_DRIVER_PATH=<path-to-sgx-driver-sources>

The path to the SGX driver sources must point to the absolute path where the SGX driver was downloaded or installed in the previous step. For example, for the DCAP version 33 of the SGX driver, you must specify ISGX_DRIVER_PATH="/usr/src/sgx-1.33/". You can define ISGX_DRIVER_PATH="" to use the default in-kernel driver’s C header.

Running make SGX=1 sgx-tokens in the test or regression directory will automatically generate the required manifest signatures (.sig files) and EINITTOKENs (.token files).

Then install graphene (recall that “direct” means non-SGX version):

meson build -Ddirect=enabled -Dsgx=enabled
ninja -C build
sudo ninja -C build install

Set -Ddirect= and -Dsgx= options to enabled or disabled according to whether you built the corresponding PAL (the snippet assumes you built both).

Additional build options

  • To create a debug build, run make DEBUG=1. This adds debug symbols in all Graphene components, builds them without optimizations, and enables detailed debug logs in Graphene.

  • To create a debug build that does not disable optimizations, run make DEBUGOPT=1.

    Note: this is generally not recommended, because optimized builds lose some debugging information, and may cause GDB to display confusing tracebacks or garbage data. You should use DEBUGOPT=1 only if you have a good reason (e.g. for profiling).

  • To build with -Werror, run make WERROR=1 and meson build --werror.

  • To specify custom mirrors for downloading the Glibc source, use make GLIBC_MIRRORS=....

  • To install into some other place than /usr/local, use meson build --prefix=<prefix>. Note that you then need to include the <prefix>/bin directory in $PATH and <prefix>/lib/python<version>/site-packages in $PYTHONPATH.