1. SDK Overview¶
Keystone SDK is work in progress, so programming enclaves requires a lot of manual efforts. However, it still provides enough libraries to run simple enclaves such as the Keystone Demo.
The Keystone SDK provides the tools required to build basic enclave hosting applications (hosts) and enclave applications (eapps).
The SDK consists of 4 main components, and examples.
- Host libraries (lib/host)
- Enclave Application libraries (lib/app)
- Edge libraries (lib/edge)
- Runtimes (rts/)
1.1. Host Libraries¶
The host libraries provide an interface for managing enclave applications via the Keystone class. Most of this library will work regardless of the runtime, but is tied directly to the kernel driver provided in the riscv-linux repository branches.
1.2. eapp Libraries¶
The eapp libraries provide both simple enclave tools (EXIT, etc) as well as some basic libc-style functions (malloc, string headers, etc).
1.3. Edge Libraries¶
The edge libraries provide features to both eapps and hosts for managing edge calls. Edge calls are function calls that cross the enclave-host boundary. Currently we only support calls from enclave->host. You can emulate host->enclave calls via polling on this interface. The edge libraries are used in many places, including the runtime and both host and eapp libraries.
The runtime is the system level code that runs in the enclave. It handles the enclave entry point and basic system calls for the userland enclave, as well as all edgecall data transfers.
2. Writing A Simple Application¶
The best way to see how to write a simple application is to look a the untrusted test eapp, and the keystone-demo repository.
A full system for using a Keystone enclave consists of possibly writing 3 things:
- Host (userspace, outside enclave, untrusted)
- Runtime (system level, inside enclave, trusted) - most users will not modify this
- Enclave app (userspace, inside enclave, trusted)
Most users will only need to write a simple Host, and use the Eyrie runtime. The bulk of the work is done in the enclave application, and in the glue that holds the components together.
NOTE: Right now all “edge” components (that is, anything that handles data and calls between enclave code and non-enclave code) is hand-written. A major next step for the SDK is to build a set of tools and compilers to do the majority of this code generation automatically.
2.1. Writing a host¶
Most host functionality is contained in the Keystone C++ class in the host library. To start an enclave application, first create one:
Keystone enclave; Params params; [... Optional params settings ...] enclave.init(PATH_TO_EAPP_ELF, PATH_TO_RUNTIME, params); edge_init(&enclave);
You can also set other enclave options via the params object, including stack size and shared memory size. edge_init is an edge wrapper function that registers edge call endpoints in the Keystone driver.
Then you start the enclave and transfer control with:
At this point, control is transferred to the enclave runtime, and then application (if using the default runtime). Generally the first action after this would be to have the enclave send its attestation report to the host. Control will be transferred back to the host code either when the enclave exits, or when an edgecall is made.
2.2. Writing an eapp¶
A generic RISC-V statically compiled ELF binary will run.
As with the host, the first action should be:
Which is another edge call wrapper function to register call points and setup buffers. Once complete, the first action should be to make an edgecall to the host to present the attestation report.
After this point, all functionality is up to the application developer. See the keystone_demo repository for an example application.