virtcontainers: constrain docker container when sandbox_cgroup_only=true

The sandbox cgroup will be constrained if there is no container
type annotation, otherwise kata will rely on container engine's cgroup
configuration

Depends-on: github.com/kata-containers/tests#2255

fixes #2408

Signed-off-by: Julio Montes <[email protected]>

cli/config/configuration-acrn.toml.in

@@ -223,9 +223,9 @@ disable_guest_seccomp=@DEFDISABLEGUESTSECCOMP@
 
 # if enabled, the runtime will add all the kata processes inside one dedicated cgroup.
 # The container cgroups in the host are not created, just one single cgroup per sandbox.
-# The sandbox cgroup is not constrained by the runtime
 # The runtime caller is free to restrict or collect cgroup stats of the overall Kata sandbox.
 # The sandbox cgroup path is the parent cgroup of a container with the PodSandbox annotation.
+# The sandbox cgroup is constrained if there is no container type annotation.
 # See: https://godoc.org/github.com/kata-containers/runtime/virtcontainers#ContainerType
 sandbox_cgroup_only=@DEFSANDBOXCGROUPONLY@
 

cli/config/configuration-clh.toml.in

@@ -199,9 +199,9 @@ disable_guest_seccomp=@DEFDISABLEGUESTSECCOMP@
 
 # if enabled, the runtime will add all the kata processes inside one dedicated cgroup.
 # The container cgroups in the host are not created, just one single cgroup per sandbox.
-# The sandbox cgroup is not constrained by the runtime
 # The runtime caller is free to restrict or collect cgroup stats of the overall Kata sandbox.
 # The sandbox cgroup path is the parent cgroup of a container with the PodSandbox annotation.
+# The sandbox cgroup is constrained if there is no container type annotation.
 # See: https://godoc.org/github.com/kata-containers/runtime/virtcontainers#ContainerType
 sandbox_cgroup_only=@DEFSANDBOXCGROUPONLY@
 

cli/config/configuration-fc.toml.in

@@ -325,9 +325,9 @@ disable_guest_seccomp=@DEFDISABLEGUESTSECCOMP@
 
 # if enable, the runtime will add all the kata processes inside one dedicated cgroup.
 # The container cgroups in the host are not created, just one single cgroup per sandbox.
-# The sandbox cgroup is not constrained by the runtime
 # The runtime caller is free to restrict or collect cgroup stats of the overall Kata sandbox.
 # The sandbox cgroup path is the parent cgroup of a container with the PodSandbox annotation.
+# The sandbox cgroup is constrained if there is no container type annotation.
 # See: https://godoc.org/github.com/kata-containers/runtime/virtcontainers#ContainerType
 sandbox_cgroup_only=@DEFSANDBOXCGROUPONLY@
 

cli/config/configuration-qemu-virtiofs.toml.in

@@ -427,9 +427,9 @@ disable_guest_seccomp=@DEFDISABLEGUESTSECCOMP@
 
 # if enabled, the runtime will add all the kata processes inside one dedicated cgroup.
 # The container cgroups in the host are not created, just one single cgroup per sandbox.
-# The sandbox cgroup is not constrained by the runtime
 # The runtime caller is free to restrict or collect cgroup stats of the overall Kata sandbox.
 # The sandbox cgroup path is the parent cgroup of a container with the PodSandbox annotation.
+# The sandbox cgroup is constrained if there is no container type annotation.
 # See: https://godoc.org/github.com/kata-containers/runtime/virtcontainers#ContainerType
 sandbox_cgroup_only=@DEFSANDBOXCGROUPONLY@
 

cli/config/configuration-qemu.toml.in

@@ -422,9 +422,9 @@ disable_guest_seccomp=@DEFDISABLEGUESTSECCOMP@
 
 # if enabled, the runtime will add all the kata processes inside one dedicated cgroup.
 # The container cgroups in the host are not created, just one single cgroup per sandbox.
-# The sandbox cgroup is not constrained by the runtime
 # The runtime caller is free to restrict or collect cgroup stats of the overall Kata sandbox.
 # The sandbox cgroup path is the parent cgroup of a container with the PodSandbox annotation.
+# The sandbox cgroup is constrained if there is no container type annotation.
 # See: https://godoc.org/github.com/kata-containers/runtime/virtcontainers#ContainerType
 sandbox_cgroup_only=@DEFSANDBOXCGROUPONLY@
 

virtcontainers/pkg/oci/utils.go

@@ -818,6 +818,8 @@ func SandboxConfig(ocispec specs.Spec, runtime RuntimeConfig, bundlePath, cid, c
 		// Spec: &ocispec,
 
 		Experimental: runtime.Experimental,
+
+		HasCRIContainerType: HasCRIContainerType(ocispec.Annotations),
 	}
 
 	if err := addAnnotations(ocispec, &sandboxConfig); err != nil {
@@ -986,3 +988,14 @@ func GetOCIConfig(status vc.ContainerStatus) (specs.Spec, error) {
 
 	return *status.Spec, nil
 }
+
+// HasCRIContainerType returns true if annottations contain
+// a CRI container type annotation
+func HasCRIContainerType(annotations map[string]string) bool {
+	for _, key := range CRIContainerTypeKeyList {
+		if _, ok := annotations[key]; ok {
+			return true
+		}
+	}
+	return false
+}

virtcontainers/sandbox.go

@@ -122,6 +122,9 @@ type SandboxConfig struct {
 
 	DisableGuestSeccomp bool
 
+	// HasCRIContainerType specifies whether container type was set explicitly through annotations or not.
+	HasCRIContainerType bool
+
 	// Experimental features enabled
 	Experimental []exp.Feature
 
@@ -2044,28 +2047,49 @@ func (s *Sandbox) setupSandboxCgroup() error {
 		return nil
 	}
 
+	s.Logger().WithField("hasCRIContainerType", s.config.HasCRIContainerType).Debug("Setting sandbox cgroup")
+
 	s.state.CgroupPath, err = validCgroupPath(spec.Linux.CgroupsPath, s.config.SystemdCgroup)
 	if err != nil {
 		return fmt.Errorf("Invalid cgroup path: %v", err)
 	}
 
-	// Do not change current cgroup configuration.
-	// Create a spec without constraints
-	unconstraintSpec := specs.Spec{
+	// Don't modify original resources, create a copy
+	resources := *spec.Linux.Resources
+	sandboxSpec := specs.Spec{
 		Linux: &specs.Linux{
-			Resources:   &specs.LinuxResources{},
-			CgroupsPath: s.state.CgroupPath,
+			Resources: &resources,
 		},
 	}
 
-	cmgr, err := newCgroupManager(s.config.Cgroups, s.state.CgroupPaths, &unconstraintSpec)
+	// kata should rely on the cgroup created and configured by
+	// container engine *only* if actual container was
+	// marked *explicitly* as sandbox through annotations.
+	if s.config.HasCRIContainerType {
+		// Do not change current cgroup configuration.
+		// Create a spec without constraints
+		sandboxSpec.Linux.Resources = &specs.LinuxResources{}
+	}
+
+	sandboxSpec.Linux.CgroupsPath = s.state.CgroupPath
+
+	// Remove this to improve device resource management, but first we need to fix some issues:
+	// - hypervisors will need access to following host's devices:
+	//   * /dev/kvm
+	//   * /dev/vhost-net
+	// - If devicemapper is the storage driver, hypervisor will need access to devicemapper devices:
+	//   * The list of cgroup devices MUST BE updated when a new container is created in the POD
+	sandboxSpec.Linux.Resources.Devices = []specs.LinuxDeviceCgroup{}
+
+	cmgr, err := newCgroupManager(s.config.Cgroups, s.state.CgroupPaths, &sandboxSpec)
 	if err != nil {
 		return fmt.Errorf("Could not create a new cgroup manager: %v", err)
 	}
 
 	runtimePid := os.Getpid()
+
 	// Add the runtime to the Kata sandbox cgroup
-	if err := cmgr.Apply(runtimePid); err != nil {
+	if err = cmgr.Apply(runtimePid); err != nil {
 		return fmt.Errorf("Could not add runtime PID %d to sandbox cgroup:  %v", runtimePid, err)
 	}
 
@@ -2078,6 +2102,10 @@ func (s *Sandbox) setupSandboxCgroup() error {
 
 	s.state.CgroupPaths = cmgr.GetPaths()
 
+	if err = cmgr.Set(&configs.Config{Cgroups: s.config.Cgroups}); err != nil {
+		return fmt.Errorf("Could not constrain cgroup: %v", err)
+	}
+
 	return nil
 }