Running Redis with Asylo

Overview

Why running in an enclave?

Enclaves are an emerging technology paradigm that protects the user application from the operating systems. Running a user application inside an enclave provides security protection for user data against attacks from the OS kernel or even a user running with root privileges.

Asylo support to run Redis

Asylo is an open source framework for developing enclave applications. Asylo provides support for full-featured applications, such as Redis 5.0.5, to run inside an enclave. We also provide an application-wrapper to make it easy to run an application in an enclave without any source code changes, along with a BUILD file that allows external sources to be built with Bazel.

Build Redis 5.0.5 with Asylo

Setting up the environment in Docker

Asylo provides a custom Docker image that contains all required dependencies, as well as Asylo’s custom toolchain, which is a convenient tool for compiling enclave applications without manually installing the toolchain. To get started, please follow the Docker instructions in the Asylo repository README.md to setup the environment for Docker.

Set up a workspace

Import Asylo and Redis

In any directory that you would like to define your project workspace, import Asylo and Redis by creating a WORKSPACE file and add the following lines to it:

workspace(name = "asylo_redis_example")

load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")

# Asylo
http_archive(
    name = "com_google_asylo",
    urls = ["https://github.com/google/asylo/archive/v0.4.0.tar.gz"],
    sha256 = "9dd8063d1a8002f6cc729f0115e2140a2eb1b14a10c111411f6b554e14ee739c",
    strip_prefix = "asylo-0.4.0",
)

# Redis
http_archive(
    name = "com_github_antirez_redis",
    build_file = "@com_google_asylo//asylo/distrib:redis.BUILD",
    urls = ["https://github.com/antirez/redis/archive/5.0.5.tar.gz"],
    sha256 = "3313d5e09938dc6e16c6911efa4a24b1c494bacb5a917a8e7925ebd55e56be85",
    strip_prefix = "redis-5.0.5",
)

load("@com_google_asylo//asylo/bazel:asylo_deps.bzl", "asylo_deps")
asylo_deps()

load("@com_google_asylo//asylo/bazel:sgx_deps.bzl", "sgx_deps")
sgx_deps()

This bazel rule imports Asylo and Redis. To build Redis with Bazel, Asylo provides the BUILD file for Redis 5.0.5, located at @com_google_asylo/asylo/distrib/redis.BUILD.

Add .bazelrc

Next, in the same workspace, create a .bazelrc file by copying from the one in Asylo Repository

This file specifies the toolchain and configurations used to build Asylo targets.

Build target for Redis 5.0.5 in SGX Simulation mode

Asylo provides an application wrapper which makes it easy to run external user applications in Asylo. To make use of it, create a BUILD file in your workspace, and add the following lines to it.

licenses(["notice"])

load("@com_google_asylo//asylo/bazel:asylo.bzl", "cc_enclave_binary")
load("@linux_sgx//:sgx_sdk.bzl", "sgx_enclave_configuration")

sgx_enclave_configuration(
    name = "redis_enclave_configuration",
    stack_max_size = "0x400000",
    heap_max_size = "0x1000000",
)

cc_enclave_binary(
    name = "asylo_redis",
    enclave_build_config = "redis_enclave_configuration",
    deps = ["@com_github_antirez_redis//:redis_main"],
)

The cc_enclave_binary is the build rule that uses the application wrapper. By claiming Redis as a dependency (redis_main is defined by our BUILD.bazel file), it will wrap Redis and run it in Asylo. The default enclave memory size is not sufficient to run Redis, so we need to also pass an sgx_enclave_configuration to increase both the stack and heap size in order to run Redis.

Run Redis Server in SGX Simulation mode

Now we are ready to build and run Redis server. First, we load a docker container that imports the current workspace by running the following Docker command from the root of your project:

docker run -it --rm \
    -v ${PWD}:/opt/my-project \
    --tmpfs /root/.cache/bazel:exec \
    -w /opt/my-project \
    --network host \
    gcr.io/asylo-framework/asylo

Here -v maps the current workspace to the directory in Docker, and -w sets the workspace in Docker. --network host allows Docker to expose ports, which is not the case by default.

As the Redis build target is created, now it can be built with the following bazel command from the Docker container:

bazel build --config=sgx-sim :asylo_redis

Specifying --config=sgx-sim builds Redis in SGX simulation mode.

After the target is built, run the following command to start Redis server:

./bazel-bin/asylo_redis

After running it you should be able to see Redis server up and messages similar as following:

5884:M 25 Mar 2019 17:08:05.024 # Server initialized
5884:M 25 Mar 2019 17:08:05.024 * Ready to accept connections

Run Redis Client

Now we can start a Redis client to connect to the server and start using it. In a different terminal, run redis-cli to connect to the server.

redis-cli

After successfully connected to the server, now you can start playing with it by setting/getting keys:

192.168.9.2:6379> ping
PONG
192.168.9.2:6379> set redis 5.0.5
OK
192.168.9.2:6379> get redis
"5.0.5"

To enable snapshotting (if it’s not automatically enabled by the config file), run from your redis client:

192.168.9.2:6379> CONFIG SET "save" "900 1"

These sets snapshotting after 900 seconds if there is at least 1 change to the dataset. You can modify the config to the snapshotting rate you would like to set.

Run Redis in SGX Hardware Mode

The following steps show how to run enclavized Redis on SGX hardware.

NOTE: The following steps only work on real SGX hardware.

Set up Enclave Config

Redis uses fork() for snapshotting. Asylo provides a secure fork() implementation that uses an authenticated Diffie-Hellman key-exchange to establish secure communication between the parent and child enclaves so that parent’s state is securely restored in a child enclave with the same identity. This security features requires configuring an SGX local assertion authority inside the enclave. The configuration is passed to the enclave via EnclaveConfig. Asylo’s cc_enclave_binary Bazel macro supports providing a custom EnclaveConfig via the application_enclave_config parameter. The following BUILD file configures the enclave with an EnclaveConfig that contains an SGX local assertion authority config.

NOTE: The security features of Asylo fork() are absent when running in simulation mode.

licenses(["notice"])

load("@com_google_asylo//asylo/bazel:asylo.bzl", "cc_enclave_binary")
load("@linux_sgx//:sgx_sdk.bzl", "sgx_enclave_configuration")

sgx_enclave_configuration(
    name = "redis_enclave_configuration",
    stack_max_size = "0x400000",
    heap_max_size = "0x1000000",
)

cc_enclave_binary(
    name = "asylo_redis",
    enclave_build_config = "redis_enclave_configuration",
    application_enclave_config = "@com_google_asylo//asylo/examples/redis:redis_enclave_config",
    deps = ["@com_github_antirez_redis//:redis_main"],
)

More details on security features provided by Asylo fork() can be found on the Asylo website.

Run Redis Server

Similar as SGX simulation case, run the following docker command from the root of your project:

docker run -it --rm \
    --device=/dev/isgx \
    -v ${PWD}:/opt/my-project \
    -v /var/run/aesmd/aesm.socket:/var/run/aesmd/aesm.socket \
    --tmpfs /root/.cache/bazel:exec \
    -w /opt/my-project \
    --network host \
    gcr.io/asylo-framework/asylo

The SGX capabilities are propagated by the docker flags --device=/dev/isgx and -v /var/run/aesmd/aesm.socket:/var/run/aesmd/aesm.socket. More details can be found on the Asylo website.

In the container shell, build the target with the following bazel command:

bazel build --config=sgx :asylo_redis

Specifying --config=sgx builds Redis in SGX hardware mode.

After the target is built, run the following command to start Redis server:

./bazel-bin/asylo_redis

After running it you should be able to see Redis server up and messages similar as following:

5884:M 25 Mar 2019 17:08:05.024 # Server initialized
5884:M 25 Mar 2019 17:08:05.024 * Ready to accept connections

Run Redis Client

The steps to connect Redis client to a server that is running in SGX hardware mode are exactly the same as in SGX simulation mode. Please follow the steps above to connect to the enclavized Redis server and set/get keys.