Store the container specs in the statefile for spec validation.

- Append all the container specs to the statefile during checkpoint.
- These specs are retrieved during restore and compare/validate with the new
specs passed during the restore.

PiperOrigin-RevId: 673550450

runsc/boot/autosave.go

@@ -56,6 +56,7 @@ func getSaveOpts(l *Loader, k *kernel.Kernel, isResume bool) state.SaveOpts {
 
 func getTargetForSaveResume(l *Loader) func(k *kernel.Kernel) {
 	return func(k *kernel.Kernel) {
+		l.addContainerSpecsToCheckpoint()
 		saveOpts := getSaveOpts(l, k, true /* isResume */)
 		// Store the state file contents in a buffer for save-resume.
 		// There is no need to verify the state file, we just need the
@@ -74,6 +75,7 @@ func getTargetForSaveRestore(l *Loader, files []*fd.FD) func(k *kernel.Kernel) {
 	var once sync.Once
 	return func(k *kernel.Kernel) {
 		once.Do(func() {
+			l.addContainerSpecsToCheckpoint()
 			saveOpts := getSaveOpts(l, k, false /* isResume */)
 			saveOpts.Destination = files[0]
 			if len(files) == 3 {

runsc/boot/loader.go

@@ -16,6 +16,7 @@
 package boot
 
 import (
+	"encoding/json"
 	"errors"
 	"fmt"
 	mrand "math/rand"
@@ -363,8 +364,16 @@ type Args struct {
 	SaveFDs []*fd.FD
 }
 
-// make sure stdioFDs are always the same on initial start and on restore
-const startingStdioFD = 256
+const (
+	// startingStdioFD is the starting stdioFD number used during sandbox
+	// start and restore. This makes sure the stdioFDs are always the same
+	// on initial start and on restore.
+	startingStdioFD = 256
+
+	// containerSpecsKey is the key used to add and pop the container specs to the
+	// kernel during save/restore.
+	containerSpecsKey = "container_specs"
+)
 
 func getRootCredentials(spec *specs.Spec, conf *config.Config, userNs *auth.UserNamespace) *auth.Credentials {
 	// Create capabilities.
@@ -1941,3 +1950,35 @@ func (l *Loader) containerRuntimeState(cid string) ContainerRuntimeState {
 	// Init process has stopped, but no one has called wait on it yet.
 	return RuntimeStateStopped
 }
+
+// addContainerSpecsToCheckpoint adds the container specs to the kernel.
+func (l *Loader) addContainerSpecsToCheckpoint() {
+	l.mu.Lock()
+	s := l.containerSpecs
+	l.mu.Unlock()
+
+	specsMap := make(map[string][]byte)
+	for k, v := range s {
+		data, err := json.Marshal(v)
+		if err != nil {
+			log.Warningf("json marshal error for specs %v", err)
+			return
+		}
+		specsMap[k] = data
+	}
+	l.k.AddStateToCheckpoint(containerSpecsKey, specsMap)
+}
+
+// popContainerSpecsFromCheckpoint pops all the container specs from the kernel.
+func popContainerSpecsFromCheckpoint(k *kernel.Kernel) (map[string]*specs.Spec, error) {
+	specsMap := (k.PopCheckpointState(containerSpecsKey)).(map[string][]byte)
+	oldSpecs := make(map[string]*specs.Spec)
+	for k, v := range specsMap {
+		var s specs.Spec
+		if err := json.Unmarshal(v, &s); err != nil {
+			return nil, fmt.Errorf("json unmarshal error for specs %v", err)
+		}
+		oldSpecs[k] = &s
+	}
+	return oldSpecs, nil
+}

runsc/boot/restore.go

@@ -137,6 +137,14 @@ func createNetworkStackForRestore(l *Loader) (*stack.Stack, inet.Stack) {
 	return nil, hostinet.NewStack()
 }
 
+// Validate OCI specs before restoring the containers.
+func validateSpecs(oldSpecs, newSpecs map[string]*specs.Spec) error {
+	if len(oldSpecs) != len(newSpecs) {
+		return fmt.Errorf("incorrect number of specs during checkpoint and restore")
+	}
+	return nil
+}
+
 func (r *restorer) restore(l *Loader) error {
 	log.Infof("Starting to restore %d containers", len(r.containers))
 
@@ -226,6 +234,14 @@ func (r *restorer) restore(l *Loader) error {
 		return err
 	}
 
+	oldSpecs, err := popContainerSpecsFromCheckpoint(l.k)
+	if err != nil {
+		return err
+	}
+	if err := validateSpecs(oldSpecs, l.containerSpecs); err != nil {
+		return err
+	}
+
 	// Since we have a new kernel we also must make a new watchdog.
 	dogOpts := watchdog.DefaultOpts
 	dogOpts.TaskTimeoutAction = l.root.conf.WatchdogAction
@@ -328,6 +344,9 @@ func (l *Loader) save(o *control.SaveOpts) (err error) {
 	}
 	o.Metadata["container_count"] = strconv.Itoa(l.containerCount())
 
+	// Save container specs.
+	l.addContainerSpecsToCheckpoint()
+
 	if err := preSaveImpl(l, o); err != nil {
 		return err
 	}