Support checkpoint/restore with pods

This adds support to checkpoint containers out of pods and restore
container into pods.

It is only possible to restore a container into a pod if it has been
checkpointed out of pod. It is also not possible to restore a non pod
container into a pod.

The main reason this does not work is the PID namespace. If a non pod
container is being restored in a pod with a shared PID namespace, at
least one process in the restored container uses PID 1 which is already
in use by the infrastructure container. If someone tries to restore
container from a pod with a shared PID namespace without a shared PID
namespace it will also fail because the resulting PID namespace will not
have a PID 1.

Signed-off-by: Adrian Reber <areber@redhat.com>

3 files changed, 148 insertions, 4 deletions

diff --git a/libpod/container_api.go b/libpod/container_api.go
index 390bba7bb..ddc5aa684 100644
--- a/libpod/container_api.go
+++ b/libpod/container_api.go
@@ -780,6 +780,16 @@ type ContainerCheckpointOptions struct {
 	// Compression tells the API which compression to use for
 	// the exported checkpoint archive.
 	Compression archive.Compression
+	// If Pod is set the container should be restored into the
+	// given Pod. If Pod is empty it is a restore without a Pod.
+	// Restoring a non Pod container into a Pod or a Pod container
+	// without a Pod is theoretically possible, but will
+	// probably not work if a PID namespace is shared.
+	// A shared PID namespace means that a Pod container has PID 1
+	// in the infrastructure container, but without the infrastructure
+	// container no PID 1 will be in the namespace and that is not
+	// possible.
+	Pod string
 }
 
 // Checkpoint checkpoints a container
@@ -811,7 +821,11 @@ func (c *Container) Checkpoint(ctx context.Context, options ContainerCheckpointO
 
 // Restore restores a container
 func (c *Container) Restore(ctx context.Context, options ContainerCheckpointOptions) error {
-	logrus.Debugf("Trying to restore container %s", c.ID())
+	if options.Pod == "" {
+		logrus.Debugf("Trying to restore container %s", c.ID())
+	} else {
+		logrus.Debugf("Trying to restore container %s into pod %s", c.ID(), options.Pod)
+	}
 	if !c.batched {
 		c.lock.Lock()
 		defer c.lock.Unlock()
diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go
index e002f9a22..bff64aa95 100644
--- a/libpod/container_internal_linux.go
+++ b/libpod/container_internal_linux.go
@@ -901,8 +901,27 @@ func (c *Container) addNamespaceContainer(g *generate.Generator, ns LinuxNS, ctr
 }
 
 func (c *Container) exportCheckpoint(options ContainerCheckpointOptions) error {
-	if len(c.Dependencies()) > 0 {
-		return errors.Errorf("Cannot export checkpoints of containers with dependencies")
+	if len(c.Dependencies()) == 1 {
+		// Check if the dependency is an infra container. If it is we can checkpoint
+		// the container out of the Pod.
+		if c.config.Pod == "" {
+			return errors.Errorf("cannot export checkpoints of containers with dependencies")
+		}
+
+		pod, err := c.runtime.state.Pod(c.config.Pod)
+		if err != nil {
+			return errors.Wrapf(err, "container %s is in pod %s, but pod cannot be retrieved", c.ID(), c.config.Pod)
+		}
+		infraID, err := pod.InfraContainerID()
+		if err != nil {
+			return errors.Wrapf(err, "cannot retrieve infra container ID for pod %s", c.config.Pod)
+		}
+		if c.Dependencies()[0] != infraID {
+			return errors.Errorf("cannot export checkpoints of containers with dependencies")
+		}
+	}
+	if len(c.Dependencies()) > 1 {
+		return errors.Errorf("cannot export checkpoints of containers with dependencies")
 	}
 	logrus.Debugf("Exporting checkpoint image of container %q to %q", c.ID(), options.TargetFile)
 
@@ -1136,10 +1155,20 @@ func (c *Container) importPreCheckpoint(input string) error {
 }
 
 func (c *Container) restore(ctx context.Context, options ContainerCheckpointOptions) (retErr error) {
-	if err := c.checkpointRestoreSupported(criu.MinCriuVersion); err != nil {
+	minCriuVersion := func() int {
+		if options.Pod == "" {
+			return criu.MinCriuVersion
+		}
+		return criu.PodCriuVersion
+	}()
+	if err := c.checkpointRestoreSupported(minCriuVersion); err != nil {
 		return err
 	}
 
+	if options.Pod != "" && !crutils.CRRuntimeSupportsPodCheckpointRestore(c.ociRuntime.Path()) {
+		return errors.Errorf("runtime %s does not support pod restore", c.ociRuntime.Path())
+	}
+
 	if !c.ensureState(define.ContainerStateConfigured, define.ContainerStateExited) {
 		return errors.Wrapf(define.ErrCtrStateInvalid, "container %s is running or paused, cannot restore", c.ID())
 	}
@@ -1247,6 +1276,83 @@ func (c *Container) restore(ctx context.Context, options ContainerCheckpointOpti
 		}
 	}
 
+	if options.Pod != "" {
+		// Running in a Pod means that we have to change all namespace settings to
+		// the ones from the infrastructure container.
+		pod, err := c.runtime.LookupPod(options.Pod)
+		if err != nil {
+			return errors.Wrapf(err, "pod %q cannot be retrieved", options.Pod)
+		}
+
+		infraContainer, err := pod.InfraContainer()
+		if err != nil {
+			return errors.Wrapf(err, "cannot retrieved infra container from pod %q", options.Pod)
+		}
+
+		infraContainer.lock.Lock()
+		if err := infraContainer.syncContainer(); err != nil {
+			infraContainer.lock.Unlock()
+			return errors.Wrapf(err, "Error syncing infrastructure container %s status", infraContainer.ID())
+		}
+		if infraContainer.state.State != define.ContainerStateRunning {
+			if err := infraContainer.initAndStart(ctx); err != nil {
+				infraContainer.lock.Unlock()
+				return errors.Wrapf(err, "Error starting infrastructure container %s status", infraContainer.ID())
+			}
+		}
+		infraContainer.lock.Unlock()
+
+		if c.config.IPCNsCtr != "" {
+			nsPath, err := infraContainer.namespacePath(IPCNS)
+			if err != nil {
+				return errors.Wrapf(err, "cannot retrieve IPC namespace path for Pod %q", options.Pod)
+			}
+			if err := g.AddOrReplaceLinuxNamespace(string(spec.IPCNamespace), nsPath); err != nil {
+				return err
+			}
+		}
+
+		if c.config.NetNsCtr != "" {
+			nsPath, err := infraContainer.namespacePath(NetNS)
+			if err != nil {
+				return errors.Wrapf(err, "cannot retrieve network namespace path for Pod %q", options.Pod)
+			}
+			if err := g.AddOrReplaceLinuxNamespace(string(spec.NetworkNamespace), nsPath); err != nil {
+				return err
+			}
+		}
+
+		if c.config.PIDNsCtr != "" {
+			nsPath, err := infraContainer.namespacePath(PIDNS)
+			if err != nil {
+				return errors.Wrapf(err, "cannot retrieve PID namespace path for Pod %q", options.Pod)
+			}
+			if err := g.AddOrReplaceLinuxNamespace(string(spec.PIDNamespace), nsPath); err != nil {
+				return err
+			}
+		}
+
+		if c.config.UTSNsCtr != "" {
+			nsPath, err := infraContainer.namespacePath(UTSNS)
+			if err != nil {
+				return errors.Wrapf(err, "cannot retrieve UTS namespace path for Pod %q", options.Pod)
+			}
+			if err := g.AddOrReplaceLinuxNamespace(string(spec.UTSNamespace), nsPath); err != nil {
+				return err
+			}
+		}
+
+		if c.config.CgroupNsCtr != "" {
+			nsPath, err := infraContainer.namespacePath(CgroupNS)
+			if err != nil {
+				return errors.Wrapf(err, "cannot retrieve Cgroup namespace path for Pod %q", options.Pod)
+			}
+			if err := g.AddOrReplaceLinuxNamespace(string(spec.CgroupNamespace), nsPath); err != nil {
+				return err
+			}
+		}
+	}
+
 	if err := c.makeBindMounts(); err != nil {
 		return err
 	}
diff --git a/libpod/oci_conmon_linux.go b/libpod/oci_conmon_linux.go
index 2914bd1a1..846d3815a 100644
--- a/libpod/oci_conmon_linux.go
+++ b/libpod/oci_conmon_linux.go
@@ -1064,6 +1064,30 @@ func (r *ConmonOCIRuntime) createOCIContainer(ctr *Container, restoreOptions *Co
 		if restoreOptions.TCPEstablished {
 			args = append(args, "--runtime-opt", "--tcp-established")
 		}
+		if restoreOptions.Pod != "" {
+			mountLabel := ctr.config.MountLabel
+			processLabel := ctr.config.ProcessLabel
+			if mountLabel != "" {
+				args = append(
+					args,
+					"--runtime-opt",
+					fmt.Sprintf(
+						"--lsm-mount-context=%s",
+						mountLabel,
+					),
+				)
+			}
+			if processLabel != "" {
+				args = append(
+					args,
+					"--runtime-opt",
+					fmt.Sprintf(
+						"--lsm-profile=selinux:%s",
+						processLabel,
+					),
+				)
+			}
+		}
 	}
 
 	logrus.WithFields(logrus.Fields{