Merge pull request #5017 from AkihiroSuda/parse-cap

oci.WithPrivileged: set the current caps, not the known caps

pkg/cri/opts/spec_linux.go

@@ -354,7 +354,7 @@ func WithDevices(osi osinterface.OS, config *runtime.ContainerConfig) oci.SpecOp
 }
 
 // WithCapabilities sets the provided capabilities from the security context
-func WithCapabilities(sc *runtime.LinuxContainerSecurityContext) oci.SpecOpts {
+func WithCapabilities(sc *runtime.LinuxContainerSecurityContext, allCaps []string) oci.SpecOpts {
 	capabilities := sc.GetCapabilities()
 	if capabilities == nil {
 		return nullOpt
@@ -366,7 +366,7 @@ func WithCapabilities(sc *runtime.LinuxContainerSecurityContext) oci.SpecOpts {
 	// AddCapabilities: []string{"ALL"}, DropCapabilities: []string{"CHOWN"}
 	// will be all capabilities without `CAP_CHOWN`.
 	if util.InStringSlice(capabilities.GetAddCapabilities(), "ALL") {
-		opts = append(opts, oci.WithAllCapabilities)
+		opts = append(opts, oci.WithCapabilities(allCaps))
 	}
 	if util.InStringSlice(capabilities.GetDropCapabilities(), "ALL") {
 		opts = append(opts, oci.WithCapabilities(nil))

pkg/cri/server/container_create_linux.go

@@ -217,10 +217,12 @@ func (c *criService) containerSpec(
 			specOpts = append(specOpts, oci.WithHostDevices, oci.WithAllDevicesAllowed)
 		} else {
 			// add requested devices by the config as host devices are not automatically added
-			specOpts = append(specOpts, customopts.WithDevices(c.os, config), customopts.WithCapabilities(securityContext))
+			specOpts = append(specOpts, customopts.WithDevices(c.os, config),
+				customopts.WithCapabilities(securityContext, c.allCaps))
 		}
 	} else { // not privileged
-		specOpts = append(specOpts, customopts.WithDevices(c.os, config), customopts.WithCapabilities(securityContext))
+		specOpts = append(specOpts, customopts.WithDevices(c.os, config),
+			customopts.WithCapabilities(securityContext, c.allCaps))
 	}
 
 	// Clear all ambient capabilities. The implication of non-root + caps

pkg/cri/server/container_create_linux_test.go

@@ -39,6 +39,7 @@ import (
 	"github.com/stretchr/testify/require"
 	runtime "k8s.io/cri-api/pkg/apis/runtime/v1alpha2"
 
+	"github.com/containerd/containerd/pkg/cap"
 	"github.com/containerd/containerd/pkg/cri/annotations"
 	"github.com/containerd/containerd/pkg/cri/config"
 	"github.com/containerd/containerd/pkg/cri/opts"
@@ -191,6 +192,7 @@ func TestContainerCapabilities(t *testing.T) {
 	testSandboxID := "sandbox-id"
 	testContainerName := "container-name"
 	testPid := uint32(1234)
+	allCaps := cap.Known()
 	for desc, test := range map[string]struct {
 		capability *runtime.Capability
 		includes   []string
@@ -208,20 +210,20 @@ func TestContainerCapabilities(t *testing.T) {
 			capability: &runtime.Capability{
 				AddCapabilities: []string{"ALL"},
 			},
-			includes: oci.GetAllCapabilities(),
+			includes: allCaps,
 		},
 		"should be able to drop all capabilities": {
 			capability: &runtime.Capability{
 				DropCapabilities: []string{"ALL"},
 			},
-			excludes: oci.GetAllCapabilities(),
+			excludes: allCaps,
 		},
 		"should be able to drop capabilities with add all": {
 			capability: &runtime.Capability{
 				AddCapabilities:  []string{"ALL"},
 				DropCapabilities: []string{"CHOWN"},
 			},
-			includes: util.SubtractStringSlice(oci.GetAllCapabilities(), "CAP_CHOWN"),
+			includes: util.SubtractStringSlice(allCaps, "CAP_CHOWN"),
 			excludes: []string{"CAP_CHOWN"},
 		},
 		"should be able to add capabilities with drop all": {
@@ -230,13 +232,14 @@ func TestContainerCapabilities(t *testing.T) {
 				DropCapabilities: []string{"ALL"},
 			},
 			includes: []string{"CAP_SYS_ADMIN"},
-			excludes: util.SubtractStringSlice(oci.GetAllCapabilities(), "CAP_SYS_ADMIN"),
+			excludes: util.SubtractStringSlice(allCaps, "CAP_SYS_ADMIN"),
 		},
 	} {
 		t.Logf("TestCase %q", desc)
 		containerConfig, sandboxConfig, imageConfig, specCheck := getCreateContainerTestData()
 		ociRuntime := config.Runtime{}
 		c := newTestCRIService()
+		c.allCaps = allCaps
 
 		containerConfig.Linux.SecurityContext.Capabilities = test.capability
 		spec, err := c.containerSpec(testID, testSandboxID, testPid, "", testContainerName, testImageName, containerConfig, sandboxConfig, imageConfig, nil, ociRuntime)

pkg/cri/server/service.go

@@ -101,6 +101,9 @@ type criService struct {
 	cniNetConfMonitor *cniNetConfSyncer
 	// baseOCISpecs contains cached OCI specs loaded via `Runtime.BaseRuntimeSpec`
 	baseOCISpecs map[string]*oci.Spec
+	// allCaps is the list of the capabilities.
+	// When nil, parsed from CapEff of /proc/self/status.
+	allCaps []string // nolint
 }
 
 // NewCRIService returns a new instance of CRIService

pkg/cri/server/service_linux.go

@@ -17,6 +17,7 @@
 package server
 
 import (
+	"github.com/containerd/containerd/pkg/cap"
 	"github.com/containerd/containerd/sys"
 	cni "github.com/containerd/go-cni"
 	"github.com/opencontainers/selinux/go-selinux"
@@ -61,6 +62,13 @@ func (c *criService) initPlatform() error {
 		return errors.Wrap(err, "failed to initialize cni")
 	}
 
+	if c.allCaps == nil {
+		c.allCaps, err = cap.Current()
+		if err != nil {
+			return errors.Wrap(err, "failed to get caps")
+		}
+	}
+
 	return nil
 }