sgx: add new special resources for TDX QGS and SGX platform registration

Signed-off-by: Mikko Ylinen <[email protected]>

Author: mythi

.github/workflows/lib-build.yaml

@@ -33,6 +33,7 @@ jobs:
           - accel-config-demo
           - intel-opencl-icd
           - openssl-qat-engine
+          - sgx-dcap-infra
           - sgx-sdk-demo
           - sgx-aesmd-demo
           - dsa-dpdk-dmadevtest

cmd/sgx_plugin/sgx_plugin.go

@@ -25,6 +25,7 @@ import (
 	dpapi "github.com/intel/intel-device-plugins-for-kubernetes/pkg/deviceplugin"
 	"k8s.io/klog/v2"
 	pluginapi "k8s.io/kubelet/pkg/apis/deviceplugin/v1beta1"
+	cdispec "tags.cncf.io/container-device-interface/specs-go"
 )
 
 const (
@@ -38,18 +39,20 @@ const (
 )
 
 type devicePlugin struct {
-	scanDone   chan bool
-	devfsDir   string
-	nEnclave   uint
-	nProvision uint
+	scanDone           chan bool
+	devfsDir           string
+	nEnclave           uint
+	nProvision         uint
+	dcapInfraResources bool
 }
 
-func newDevicePlugin(devfsDir string, nEnclave, nProvision uint) *devicePlugin {
+func newDevicePlugin(devfsDir string, nEnclave, nProvision uint, dcapInfraResources bool) *devicePlugin {
 	return &devicePlugin{
-		devfsDir:   devfsDir,
-		nEnclave:   nEnclave,
-		nProvision: nProvision,
-		scanDone:   make(chan bool, 1),
+		devfsDir:           devfsDir,
+		nEnclave:           nEnclave,
+		nProvision:         nProvision,
+		dcapInfraResources: dcapInfraResources,
+		scanDone:           make(chan bool, 1),
 	}
 }
 
@@ -96,6 +99,39 @@ func (dp *devicePlugin) scan() (dpapi.DeviceTree, error) {
 		devTree.AddDevice(deviceTypeProvision, devID, dpapi.NewDeviceInfoWithTopologyHints(pluginapi.Healthy, nodes, nil, nil, nil, nil, nil))
 	}
 
+	if !dp.dcapInfraResources {
+		return devTree, nil
+	}
+
+	tdQeNodes := []pluginapi.DeviceSpec{
+		{HostPath: sgxEnclavePath, ContainerPath: sgxEnclavePath, Permissions: "rw"},
+		{HostPath: sgxProvisionPath, ContainerPath: sgxProvisionPath, Permissions: "rw"},
+	}
+
+	devTree.AddDevice("tdqe", "tdqe-1", dpapi.NewDeviceInfoWithTopologyHints(pluginapi.Healthy, tdQeNodes, nil, nil, nil, nil, nil))
+
+	regNodes := []pluginapi.DeviceSpec{
+		{HostPath: sgxEnclavePath, ContainerPath: sgxEnclavePath, Permissions: "rw"},
+	}
+
+	// TODO: /sys/firmware is a maskedPath. Test /run/efivars with a patched PCK-ID-Retrieval-Tool.
+	efiVarFsMount := &cdispec.Spec{
+		Version: dpapi.CDIVersion,
+		Kind:    dpapi.CDIVendor + "/sgx",
+		Devices: []cdispec.Device{
+			{
+				Name: "efivarfs",
+				ContainerEdits: cdispec.ContainerEdits{
+					Mounts: []*cdispec.Mount{
+						{HostPath: "efivarfs", ContainerPath: "/run/efivars", Type: "efivarfs", Options: []string{"rw", "nosuid", "nodev", "noexec", "relatime"}},
+					},
+				},
+			},
+		},
+	}
+
+	devTree.AddDevice("registration", "registration-1", dpapi.NewDeviceInfoWithTopologyHints(pluginapi.Healthy, regNodes, nil, nil, nil, nil, efiVarFsMount))
+
 	return devTree, nil
 }
 
@@ -121,15 +157,18 @@ func getDefaultPodCount(nCPUs uint) uint {
 func main() {
 	var enclaveLimit, provisionLimit uint
 
+	var dcapInfraResources bool
+
 	podCount := getDefaultPodCount(uint(runtime.NumCPU()))
 
 	flag.UintVar(&enclaveLimit, "enclave-limit", podCount, "Number of \"enclave\" resources")
 	flag.UintVar(&provisionLimit, "provision-limit", podCount, "Number of \"provision\" resources")
+	flag.BoolVar(&dcapInfraResources, "dcap-infra-resources", false, "add special resources for DCAP infrastructure daemonSet pods")
 	flag.Parse()
 
 	klog.V(4).Infof("SGX device plugin started with %d \"%s/enclave\" resources and %d \"%s/provision\" resources.", enclaveLimit, namespace, provisionLimit, namespace)
 
-	plugin := newDevicePlugin(devicePath, enclaveLimit, provisionLimit)
+	plugin := newDevicePlugin(devicePath, enclaveLimit, provisionLimit, dcapInfraResources)
 	manager := dpapi.NewManager(namespace, plugin)
 	manager.Run()
 }

cmd/sgx_plugin/sgx_plugin_test.go

@@ -27,17 +27,22 @@ func init() {
 	_ = flag.Set("v", "4") // Enable debug output
 }
 
+// Update if new resource types are added.
+const dcapInfraResources = 2
+
 // mockNotifier implements Notifier interface.
 type mockNotifier struct {
 	scanDone          chan bool
 	enclaveDevCount   int
 	provisionDevCount int
+	dcapInfraResCnt   int
 }
 
 // Notify stops plugin Scan.
 func (n *mockNotifier) Notify(newDeviceTree dpapi.DeviceTree) {
 	n.enclaveDevCount = len(newDeviceTree[deviceTypeEnclave])
 	n.provisionDevCount = len(newDeviceTree[deviceTypeProvision])
+	n.dcapInfraResCnt = len(newDeviceTree) - n.enclaveDevCount - n.provisionDevCount
 	n.scanDone <- true
 }
 
@@ -95,6 +100,7 @@ func TestScan(t *testing.T) {
 		requestedProvisionDevs uint
 		expectedEnclaveDevs    int
 		expectedProvisionDevs  int
+		requestDcapInfra       bool
 	}{
 		{
 			name: "no device installed",
@@ -131,6 +137,16 @@ func TestScan(t *testing.T) {
 			requestedProvisionDevs: 20,
 			expectedProvisionDevs:  20,
 		},
+		{
+			name:                   "all resources",
+			enclaveDevice:          "sgx_enclave",
+			provisionDevice:        "sgx_provision",
+			requestedEnclaveDevs:   1,
+			expectedEnclaveDevs:    1,
+			requestedProvisionDevs: 1,
+			expectedProvisionDevs:  1,
+			requestDcapInfra:       true,
+		},
 	}
 
 	for _, tc := range tcases {
@@ -159,7 +175,7 @@ func TestScan(t *testing.T) {
 				}
 			}
 
-			plugin := newDevicePlugin(devfs, tc.requestedEnclaveDevs, tc.requestedProvisionDevs)
+			plugin := newDevicePlugin(devfs, tc.requestedEnclaveDevs, tc.requestedProvisionDevs, tc.requestDcapInfra)
 
 			notifier := &mockNotifier{
 				scanDone: plugin.scanDone,
@@ -175,6 +191,9 @@ func TestScan(t *testing.T) {
 			if tc.expectedProvisionDevs != notifier.provisionDevCount {
 				t.Errorf("Wrong number of discovered provision devices")
 			}
+			if tc.requestDcapInfra && notifier.dcapInfraResCnt != dcapInfraResources {
+				t.Errorf("Wrong number of discovered DCAP infra resources: expected %d, got %d.", dcapInfraResources, notifier.dcapInfraResCnt)
+			}
 		})
 	}
 }

demo/sgx-dcap-infra/Dockerfile

@@ -0,0 +1,22 @@
+FROM ubuntu:24.04
+
+# TODO: pin DCAP release version
+
+RUN apt update && apt install -y curl gnupg \
+    && echo "deb [arch=amd64 signed-by=/usr/share/keyrings/intel-sgx.gpg] https://download.01.org/intel-sgx/sgx_repo/ubuntu noble main" | \
+    tee -a /etc/apt/sources.list.d/intel-sgx.list \
+    && curl -s https://download.01.org/intel-sgx/sgx_repo/ubuntu/intel-sgx-deb.key | \
+    gpg --dearmor --output /usr/share/keyrings/intel-sgx.gpg \
+    && apt update \
+    && apt install -y --no-install-recommends \
+       tdx-qgs \
+       sgx-pck-id-retrieval-tool \
+       libsgx-ra-uefi \
+       libsgx-dcap-default-qpl
+
+# BUG: "qgs -p=0" gets overriden by the config file making the parameter useless
+RUN sed -e 's/\(^port =\).*/\1 0/g' -i /etc/qgs.conf
+
+COPY dcap-registration-flow /usr/bin
+
+ENTRYPOINT ["/opt/intel/tdx-qgs/qgs", "--no-daemon", "-p=0"]

demo/sgx-dcap-infra/dcap-registration-flow

@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+
+set -u
+
+# TODO remove before merging
+sleep infinity
+
+if [ ! -x "${PWD}"/PCKIDRetrievalTool ]; then
+	echo "dcap-registration-flow: PCKIDRetrievalTool must be in the workingDir and executable"
+	exit 1
+fi
+
+echo "Waiting for the PCCS to be ready ..."
+
+if ! curl --retry 20 --retry-delay 30 -k https://pccs-service:8042/sgx/certification/v4/rootcacrl &> /dev/null; then
+	echo "ERROR: PCCS pod didn't become ready after 20 minutes"
+	exit 1
+fi
+
+echo "PCCS is online, proceeding ..."
+
+ARGS="-user_token ${USER_TOKEN} -url ${PCCS_URL} -use_secure_cert ${SECURE_CERT}"
+
+echo "Calling PCKIDRetrievalTool ${ARGS} ..."
+
+./PCKIDRetrievalTool ${ARGS}
+
+sleep infinity

deployments/operator/crd/bases/deviceplugin.intel.com_sgxdeviceplugins.yaml

@@ -55,6 +55,10 @@ spec:
           spec:
             description: SgxDevicePluginSpec defines the desired state of SgxDevicePlugin.
             properties:
+              dcapInfraResources:
+                description: DcapInfraResources adds two special resources for DCAP
+                  infra DaemonSet Pods.
+                type: boolean
               enclaveLimit:
                 description: EnclaveLimit is a number of containers that can share
                   the same SGX enclave device.

deployments/sgx_dcap/base/intel-sgx-dcap.yaml

@@ -0,0 +1,46 @@
+apiVersion: apps/v1
+kind: DaemonSet
+metadata:
+  name: intel-sgx-dcap-infra
+  labels:
+    app: intel-sgx-dcap-infra
+spec:
+  selector:
+    matchLabels:
+      app: intel-sgx-dcap-infra
+  template:
+    metadata:
+      labels:
+        app: intel-sgx-dcap-infra
+    spec:
+      automountServiceAccountToken: false
+      initContainers:
+      - name: platform-registration
+        image: intel/sgx-dcap-infra:devel
+        restartPolicy: Always
+        workingDir: "/opt/intel/sgx-pck-id-retrieval-tool/"
+        command: ['/usr/bin/dcap-registration-flow']
+        securityContext:
+          readOnlyRootFilesystem: true
+          allowPrivilegeEscalation: false
+        resources:
+          limits:
+            sgx.intel.com/registration: 1
+      containers:
+      - name: tdxqgs
+        image: intel/sgx-dcap-infra:devel
+        securityContext:
+          readOnlyRootFilesystem: true
+          allowPrivilegeEscalation: false
+        resources:
+          limits:
+            sgx.intel.com/tdqe: 1
+        imagePullPolicy: IfNotPresent
+        volumeMounts:
+        - name: qgs-socket
+          mountPath: /var/run/tdx-qgs
+      volumes:
+      - name: qgs-socket
+        hostPath:
+          path: /var/run/tdx-qgs
+          type: DirectoryOrCreate

deployments/sgx_dcap/base/kustomization.yaml

@@ -0,0 +1,4 @@
+resources:
+- intel-sgx-dcap.yaml
+generatorOptions:
+  disableNameSuffixHash: true

deployments/sgx_dcap/kustomization.yaml

@@ -0,0 +1,2 @@
+resources:
+  - base

deployments/sgx_plugin/base/intel-sgx-plugin.yaml

@@ -49,6 +49,8 @@ spec:
         - name: sgx-provision
           mountPath: /dev/sgx_provision
           readOnly: true
+        - name: cdipath
+          mountPath: /var/run/cdi
       volumes:
       - name: kubeletsockets
         hostPath:
@@ -61,5 +63,9 @@ spec:
         hostPath:
           path: /dev/sgx_provision
           type: CharDevice
+      - name: cdipath
+        hostPath:
+          path: /var/run/cdi
+          type: DirectoryOrCreate
       nodeSelector:
         kubernetes.io/arch: amd64