Rust SGX SDK

Rust SGX SDK helps developers write Intel SGX applications in Rust programming language. [Paper pdf]

v0.9.7 Release

This version provides a new namespace: sgx_tstd::untrusted, including sgx_tstd::untrusted::fs sgx_tstd::untrusted::time and sgx_tstd::untrusted::path, providing supports to operation to ocalls in a untrusted namespace. The untrusted namespace is always enabled no matter untrusted_* is set or not. We urge the developers to use the sgx_tstd::untrusted namespace to port their crates, instead of enabling the untrusted_ series of features. Also, we renamed the untrusted_net feature to net for feature name unification. Please refer to release_notes for further details.

v0.9.6 Release

This version provides security enhancement for untrusted IO and additional support for monotonic counter. Untrusted IO operations in sgx_tstd::fs sgx_tstd::net and sgx_tstd::time are DISABLED by default to reduce the untrusted surface, and can be enabled by features. Trusted time support is moved to sgx_tservice::sgxtime and monotonic counter is provided by sgx_tservice::sgxcounter. Please refer to release_notes for further details.

v0.9.5 Release

This is a milestone version, and may be the last version before 1.0.0. It provides supports of network access, TLS connection, trusted/untrusted file system access, trusted/untrusted time, and environment variable operations. Most important, it supports xargo! Now x86_64-unknown-linux-sgx is the new platform target. All of the code samples and third-party libraries could be built by xargo via XARGO_SGX=1 make (cargo also supported by make). What's more, we provide a pair of TLS client/server sample as a complete stack of secure! Please refer to release_notes for further details.

Run Rust SGX applications in Mesalock Linux

MesaLock Linux is a general purpose Linux distribution which aims to provide a safe and secure user space environment. Now we can run Rust SGX applications in Mesalock Linux within a few steps. Please refer to the tutorial for details.

v0.9.1 Release

This version supports the recently released Intel SGX SDK 2.0, and provides the most popular machine learning library rusty-machine in Intel SGX! Please refer to release_notes for further details.

v0.9.0 Release

Almost there! Rust SGX SDK v0.9.0 is coming up as a beta version of the future v1.0.0, with the most desired sgx::tstd as well as many new features! Also we added support for programming SGX untrusted part in Rust using sgx::urts. Now it's easy to port Rust crates to the SGX trust execution environment and write the whole SGX application in Rust! Please refer to release_notes for further details.

Good news! Our poster 'Rust SGX SDK: Towards Memory Safety in Intel SGX Enclave' [pdf] has been accepted by CCS'17. Please kindly cite our poster if you like Rust SGX SDK!

v0.2.0 Release

We are proud to have our v0.2.0 Rust SGX SDK released. It is now providing more threading functions, thread local storages, exception handling routines and supports unwind mechanism, as well as support of LARGE ENCLAVE MEMORY with the help of Intel SGX v1.9 (31.75 GB enclave memory tested). Please refer to release notes for further details. And we are working on a white paper for technical details about this project.

Attention Rust has recently merged a patch (rustc: Implement the #[global_allocator] attribute) which significantly changes the behavior of liballoc. Thus set_oom_handler is no longer available since nightly-2017-07-07. Due to its instability, v0.2.0 Rust SGX SDK keeps using the old liballoc instead of new liballoc_system. As a result, nightly version rustc after 2017-07-06 will not successfully compile sgx_trts.

Requirement

Ubuntu 16.04

Intel SGX SDK 2.0 for Linux installed

Docker (Recommended)

Configuration

The docker image now supports Intel ME. If you need it, please refer to the sgxtime readme for instructions.

Using docker (Recommended) without ME support

First, make sure Intel SGX Driver 2.0 is installed and functions well. /dev/isgx should be appeared.

Second, pull the docker image

$ docker pull baiduxlab/sgx-rust

Third, start a docker with sgx device support and the Rust SGX SDK.

$ docker run -v /your/path/to/rust-sgx:/root/sgx -ti --device /dev/isgx baiduxlab/sgx-rust

Next, start the aesm service inside the docker

root@docker:/# /opt/intel/sgxpsw/aesm/aesm_service &

Finally, check if the sample code works

root@docker:~/sgx/samplecode/helloworld# make

root@docker:~/sgx/samplecode/helloworld# cd bin

root@docker:~/sgx/samplecode/helloworld/bin# ./app

Native without docker (Not recommended)

Install Intel SGX driver and SDK first. And refer to Dockerfile for detail.

Build the docker image by yourself

Make sure Intel SGX SDK is properly installed and service started on the host OS. Then cd dockerfile and run docker build -t rust-sgx-docker to build.

Documents

The online documents for SDK crates can be found here.

We recommend to use cargo doc to generate documents for each crate in this SDK by yourself. The auto generated documents are easy to read and search.

Sample Codes

We provide five sample codes to help developers understand how to write Enclave codes in Rust. These codes are located at samplecode directory.

• helloworld is a very simple app. It shows some basic usages of argument passing, Rust string and ECALL/OCALLs.

• crypto shows the usage of crypto APIs provided by Intel SGX libraries. It does some crypto calculations inside the enclave, which is recommended in most circumstances.

• localattestation is a sample ported from the original Intel SGX SDK. It shows how to do local attestation in Rust programming language.

• sealeddata sample shows how to seal secret data in an enclave and how to verify the sealed data.

• thread sample is a sample ported from the original Intel SGX SDK, showing some basic usages of threading APIs.

• remoteattestation sample shows how to make remote attestation with Rust SGX SDK. The sample is forked from linux-sgx-attestation and credits to Blackrabbit ([email protected]). The enclave in Rust is shipped in this sample and Makefiles are modified accordingly.

• hugemem sample shows how to use huge mem in SGX enclave. In this sample, we allocate reserve 31.75GB heap space and allocate 31.625GB buffers!

• file sample shows how to read/write files in SGX enclave.

• hello-rust is the helloworld sample writtin in pure Rust.

• backtrace is a sample showing how to enabling backtrace mechanism inside the enclave.

• unit-test shows the way of writing unit tests and conduct unit testing.

• zlib-lazy-static-sample shows how to use ported third party crates inside enclave.

• machine-learning shows how to use rusty-machine for machine learning inside Intel SGX enclave.

• New! tls contains a pair of TLS client/server runs perfectly in SGX enclave!

• New! sgxtime shows how to acquire trusted timestamp via Intel ME. Please refer to this instruction for detail.

Samples of ported third-party libraries

As of v0.9.5, we provide 25 ported third-party libraries. All of them could be compiled using xargo (XARGO_SGX=1 make) or cargo (make).

Tips for writing enclaves in Rust

Writing EDL

• For fixed-length array in ECALL/OCALL definition, declare it as an array. For dynamic-length array, use the keyword size= to let the Intel SGX knows how many bytes should be copied.

ECALL Function Naming

• Add #[no_mangle] for every ECALL function.

Passing/returning arrays

• For dynamic-length array, the only way is to use raw pointers in Rust. There are several functions to get/set data using raw pointers such as offset method. One can also use slice::from_raw_parts to convert the array to a slice.

• For Fixed-length array, the above method is acceptable. And according to discussions in issue 30382 and issue 31227, thin-pointers (such as fixed-length array) are FFI-safe for now, but undocumented. In the sample codes, we use fixed-length arrays for passing and returning some fixed-length data.

License

Baidu Rust-SGX SDK is provided under the BSD license. Please refer to the License file for details.

Authors

Ran Duan, Long Li, Shi Jia, Yu Ding, Lenx Wei, Tanghui Chen

Acknowledgement

Thanks to Prof. Jingqiang Lin for his contribution to this project.

Contacts

Yu Ding, [email protected]