Support PID NamespaceMode_TARGET

This commit adds support for the PID namespace mode TARGET
when generating a container spec.

The container that is created will be sharing its PID namespace
with the target container that was specified by ID in the namespace
options.

Signed-off-by: Thomas Hartland <thomas.george.hartland@cern.ch>

pkg/cri/opts/spec_linux.go

@@ -593,16 +593,16 @@ func WithSupplementalGroups(groups []int64) oci.SpecOpts {
 }
 
 // WithPodNamespaces sets the pod namespaces for the container
-func WithPodNamespaces(config *runtime.LinuxContainerSecurityContext, pid uint32) oci.SpecOpts {
+func WithPodNamespaces(config *runtime.LinuxContainerSecurityContext, sandboxPid uint32, targetPid uint32) oci.SpecOpts {
 	namespaces := config.GetNamespaceOptions()
 
 	opts := []oci.SpecOpts{
-		oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.NetworkNamespace, Path: GetNetworkNamespace(pid)}),
-		oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.IPCNamespace, Path: GetIPCNamespace(pid)}),
-		oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.UTSNamespace, Path: GetUTSNamespace(pid)}),
+		oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.NetworkNamespace, Path: GetNetworkNamespace(sandboxPid)}),
+		oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.IPCNamespace, Path: GetIPCNamespace(sandboxPid)}),
+		oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.UTSNamespace, Path: GetUTSNamespace(sandboxPid)}),
 	}
 	if namespaces.GetPid() != runtime.NamespaceMode_CONTAINER {
-		opts = append(opts, oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.PIDNamespace, Path: GetPIDNamespace(pid)}))
+		opts = append(opts, oci.WithLinuxNamespace(runtimespec.LinuxNamespace{Type: runtimespec.PIDNamespace, Path: GetPIDNamespace(targetPid)}))
 	}
 	return oci.Compose(opts...)
 }

pkg/cri/server/container_create_linux.go

@@ -270,9 +270,24 @@ func (c *criService) containerSpec(
 		specOpts = append(specOpts, customopts.WithAnnotation(pKey, pValue))
 	}
 
+	// Default target PID namespace is the sandbox PID.
+	targetPid := sandboxPid
+	// If the container targets another container's PID namespace,
+	// set targetPid to the PID of that container.
+	nsOpts := securityContext.GetNamespaceOptions()
+	if nsOpts.GetPid() == runtime.NamespaceMode_TARGET {
+		targetContainer, err := c.validateTargetContainer(sandboxID, nsOpts.TargetId)
+		if err != nil {
+			return nil, errors.Wrapf(err, "invalid target container")
+		}
+
+		status := targetContainer.Status.Get()
+		targetPid = status.Pid
+	}
+
 	specOpts = append(specOpts,
 		customopts.WithOOMScoreAdj(config, c.config.RestrictOOMScoreAdj),
-		customopts.WithPodNamespaces(securityContext, sandboxPid),
+		customopts.WithPodNamespaces(securityContext, sandboxPid, targetPid),
 		customopts.WithSupplementalGroups(supplementalGroups),
 		customopts.WithAnnotation(annotations.ContainerType, annotations.ContainerTypeContainer),
 		customopts.WithAnnotation(annotations.SandboxID, sandboxID),

pkg/cri/server/container_start.go

@@ -84,6 +84,17 @@ func (c *criService) StartContainer(ctx context.Context, r *runtime.StartContain
 		return nil, errors.Errorf("sandbox container %q is not running", sandboxID)
 	}
 
+	// Recheck target container validity in Linux namespace options.
+	if linux := config.GetLinux(); linux != nil {
+		nsOpts := linux.GetSecurityContext().GetNamespaceOptions()
+		if nsOpts.GetPid() == runtime.NamespaceMode_TARGET {
+			_, err := c.validateTargetContainer(sandboxID, nsOpts.TargetId)
+			if err != nil {
+				return nil, errors.Wrap(err, "invalid target container")
+			}
+		}
+	}
+
 	ioCreation := func(id string) (_ containerdio.IO, err error) {
 		stdoutWC, stderrWC, err := c.createContainerLoggers(meta.LogPath, config.GetTty())
 		if err != nil {

pkg/cri/server/helpers.go

@@ -242,6 +242,30 @@ func (c *criService) ensureImageExists(ctx context.Context, ref string, config *
 	return &newImage, nil
 }
 
+// validateTargetContainer checks that a container is a valid
+// target for a container using PID NamespaceMode_TARGET.
+// The target container must be in the same sandbox and must be running.
+// Returns the target container for convenience.
+func (c *criService) validateTargetContainer(sandboxID, targetContainerID string) (containerstore.Container, error) {
+	targetContainer, err := c.containerStore.Get(targetContainerID)
+	if err != nil {
+		return containerstore.Container{}, errors.Wrapf(err, "container %q does not exist", targetContainerID)
+	}
+
+	targetSandboxID := targetContainer.Metadata.SandboxID
+	if targetSandboxID != sandboxID {
+		return containerstore.Container{},
+			errors.Errorf("container %q (sandbox %s) does not belong to sandbox %s", targetContainerID, targetSandboxID, sandboxID)
+	}
+
+	status := targetContainer.Status.Get()
+	if state := status.State(); state != runtime.ContainerState_CONTAINER_RUNNING {
+		return containerstore.Container{}, errors.Errorf("container %q is not running - in state %s", targetContainerID, state)
+	}
+
+	return targetContainer, nil
+}
+
 // isInCRIMounts checks whether a destination is in CRI mount list.
 func isInCRIMounts(dst string, mounts []*runtime.Mount) bool {
 	for _, m := range mounts {