Kubernetes Device Plugin for Intel SGX2/SGX1.
English | 简体中文
sgx-device-plugin
is a Kubernetes Device Plugin powered by Alibaba Cloud and Ant Financial, making it easier to run SGX1/SGX2 applications inside a container.
Intel(R) Software Guard Extensions (Intel(R) SGX) is an Intel technology for application developers seeking to protect select code and data from disclosure or modification. See official introduction for more details.
- Using SGX features without privileged mode.
- Support retrieving real SGX1/SGX2 EPC size.
- Support EPC resource allocation.
- Support SGX1(/dev/isgx, /dev/sgx) drivers passthrough.
- Support SGX2(/dev/sgx_enclave, /dev/sgx_provision, /dev/sgx/enclave, /dev/sgx/provision) drivers passthrough.
- For SGX1
- Intel SGX Drivers
- Intel SGX PSW(Platform Software) (If you need AESM)
- Kubernetes version >= 1.10
- Go version >= 1.13
TEE (Trusted Execution Environment), created by hardware isolation and memory encryption technology such as Intel SGX, is a special execution context named enclave which confidential code and data runs inside. It aims to help application owners to protect their data and prevent data steals by other applications, kernel, BIOS, even all hardware beside CPU.
You could create a confidential Kubernetes cluster using ACK (Alibaba Cloud Container Service for Kubernetes), all worker nodes are running on bare-metal sgx-enabled machines(model: ecs.ebmhfg5.2xlarge
) which have less overhead, better performance and more stable than VM. By default, containerd, Intel SGX Driver, Intel SGX PSW(Platform Software) and SGX-Device-Plugin will be installed on each node.
Step 1: Download source code and build binary.
mkdir -p $GOPATH/src/github.com/AliyunContainerService
git clone https://github.com/AliyunContainerService/sgx-device-plugin.git $GOPATH/src/github.com/AliyunContainerService/sgx-device-plugin
cd $GOPATH/src/github.com/AliyunContainerService/sgx-device-plugin/
make
ls -l _output/sgx-device-plugin
Step 2: Build Image.
docker build -t {SGX_DEVICE_PLUGIN_IMAGE} . -f Dockerfile
docker push {SGX_DEVICE_PLUGIN_IMAGE}
While you are creating a confidential Kubernetes cluster using ACK(Alibaba Cloud Container Service for Kubernetes), sgx-device-plugin will be installed by default. Also, you may install it on your own private Kubernetes cluster manually.
$ cat <<EOF | kubectl create -f -
apiVersion: apps/v1
kind: DaemonSet
metadata:
name: sgx-device-plugin-ds
namespace: kube-system
spec:
selector:
matchLabels:
k8s-app: sgx-device-plugin
template:
metadata:
annotations:
scheduler.alpha.kubernetes.io/critical-pod: ""
labels:
k8s-app: sgx-device-plugin
spec:
containers:
- image: registry.cn-hangzhou.aliyuncs.com/acs/sgx-device-plugin:v1.0.0-fb467e2-aliyun
imagePullPolicy: IfNotPresent
name: sgx-device-plugin
securityContext:
allowPrivilegeEscalation: false
capabilities:
drop:
- ALL
volumeMounts:
- mountPath: /var/lib/kubelet/device-plugins
name: device-plugin
- mountPath: /dev
name: dev
tolerations:
- effect: NoSchedule
key: alibabacloud.com/sgx_epc_MiB
operator: Exists
volumes:
- hostPath:
path: /var/lib/kubelet/device-plugins
type: DirectoryOrCreate
name: device-plugin
- hostPath:
path: /dev
type: Directory
name: dev
EOF
$ kubectl -n kube-system -l k8s-app=sgx-device-plugin
NAME READY STATUS RESTARTS AGE
sgx-device-plugin-ds-5brgs 1/1 Running 0 5d5h
sgx-device-plugin-ds-b467q 1/1 Running 0 5d5h
sgx-device-plugin-ds-vl7sm 1/1 Running 0 5d5h
$
While plugins are running, run command kubectl get node {NODE_NAME} -o yaml
, then you will find a new resource type: alibabacloud.com/sgx_epc_MiB
.
$ kubectl get node {NODE_NAME} -o yaml
...
allocatable:
alibabacloud.com/sgx_epc_MiB: "93"
cpu: "8"
...
$
Your application MUST BE SGX-enabled, means that your application is built and signed with SGX SDK, such as Intel SGX SDK.
apiVersion: v1
kind: Pod
metadata:
name: {POD_NAME}
namespace: default
spec:
containers:
- image: {CONTAINER_IMAGE}
imagePullPolicy: IfNotPresent
name: {CONTAINER_NAME}
resources:
requests:
alibabacloud.com/sgx_epc_MiB: 20
limits:
alibabacloud.com/sgx_epc_MiB: 20
If you want a remote attestation, aesm.socket MUST BE mounted inside application containers. There are two ways to achieve it:
Way 1: Mount aesm.socket (e.g. /var/run/aesmd/aesm.socket) inside your application containers manually, maybe like this:
apiVersion: v1
kind: Pod
metadata:
name: {POD_NAME}
namespace: default
spec:
containers:
- image: {CONTAINER_IMAGE}
imagePullPolicy: IfNotPresent
name: {CONTAINER_NAME}
resources:
requests:
alibabacloud.com/sgx_epc_MiB: 20
limits:
alibabacloud.com/sgx_epc_MiB: 20
volumeMounts:
- mountPath: /var/run/aesmd/aesm.socket
name: aesmsocket
volumes:
- hostPath:
path: /var/run/aesmd/aesmd/aesm.socket
type: Socket
name: aesmsocket
Way 2: Enable AESM socket attachment of sgx-device-plugin (via --enable-aesm-socket-attach=true) which will help you mount ASEM socket inside your application containers automatically. See deploy/sgx-device-plugin-enable-aesm-socket-attach.yaml.
-
Can I deploy this SGX device plugin in my own self-hosting Kubernetes?
Yes, this plugin is cloud native, you can run it on sgx-enabled nodes in any Kubernetes. -
Does this plugin actually limit EPC size for sgx-enabled container?
No, EPC size limitation specified byalibabacloud.com/sgx_epc_MiB
is just used for kube-scheduler.
Currently, SGX driver doesn't support EPC size limitation.
This software is released under the Apache 2.0 license.
See CONTRIBUTING.md for details.