Support VFS2 save/restore.

Inode number consistency checks are now skipped in save/restore tests for
reasons described in greatest detail in StatTest.StateDoesntChangeAfterRename.
They pass in VFS1 due to the bug described in new test case
SimpleStatTest.DifferentFilesHaveDifferentDeviceInodeNumberPairs.

Fixes #1663

PiperOrigin-RevId: 338776148

2 files changed, 41 insertions, 6 deletions

diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go
index 894651519..4e0f0d57a 100644
--- a/runsc/boot/controller.go
+++ b/runsc/boot/controller.go
@@ -30,6 +30,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/socket/netstack"
 	"gvisor.dev/gvisor/pkg/sentry/state"
 	"gvisor.dev/gvisor/pkg/sentry/time"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/sentry/watchdog"
 	"gvisor.dev/gvisor/pkg/tcpip/stack"
 	"gvisor.dev/gvisor/pkg/urpc"
@@ -367,12 +368,20 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error {
 	cm.l.k = k
 
 	// Set up the restore environment.
+	ctx := k.SupervisorContext()
 	mntr := newContainerMounter(cm.l.root.spec, cm.l.root.goferFDs, cm.l.k, cm.l.mountHints)
-	renv, err := mntr.createRestoreEnvironment(cm.l.root.conf)
-	if err != nil {
-		return fmt.Errorf("creating RestoreEnvironment: %v", err)
+	if kernel.VFS2Enabled {
+		ctx, err = mntr.configureRestore(ctx, cm.l.root.conf)
+		if err != nil {
+			return fmt.Errorf("configuring filesystem restore: %v", err)
+		}
+	} else {
+		renv, err := mntr.createRestoreEnvironment(cm.l.root.conf)
+		if err != nil {
+			return fmt.Errorf("creating RestoreEnvironment: %v", err)
+		}
+		fs.SetRestoreEnvironment(*renv)
 	}
-	fs.SetRestoreEnvironment(*renv)
 
 	// Prepare to load from the state file.
 	if eps, ok := networkStack.(*netstack.Stack); ok {
@@ -399,7 +408,7 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error {
 
 	// Load the state.
 	loadOpts := state.LoadOpts{Source: specFile}
-	if err := loadOpts.Load(k, networkStack, time.NewCalibratedClocks()); err != nil {
+	if err := loadOpts.Load(ctx, k, networkStack, time.NewCalibratedClocks(), &vfs.CompleteRestoreOptions{}); err != nil {
 		return err
 	}
 
diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go
index 004da5b40..b157387ef 100644
--- a/runsc/boot/vfs.go
+++ b/runsc/boot/vfs.go
@@ -210,6 +210,9 @@ func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *c
 		ReadOnly: c.root.Readonly,
 		GetFilesystemOptions: vfs.GetFilesystemOptions{
 			Data: strings.Join(data, ","),
+			InternalData: gofer.InternalFilesystemOptions{
+				UniqueID: "/",
+			},
 		},
 		InternalMount: true,
 	}
@@ -427,6 +430,7 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mo
 	fsName := m.Type
 	useOverlay := false
 	var data []string
+	var iopts interface{}
 
 	// Find filesystem name and FS specific data field.
 	switch m.Type {
@@ -451,6 +455,9 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mo
 			return "", nil, false, fmt.Errorf("9P mount requires a connection FD")
 		}
 		data = p9MountData(m.fd, c.getMountAccessType(m.Mount), true /* vfs2 */)
+		iopts = gofer.InternalFilesystemOptions{
+			UniqueID: m.Destination,
+		}
 
 		// If configured, add overlay to all writable mounts.
 		useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly
@@ -462,7 +469,8 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *config.Config, m *mo
 
 	opts := &vfs.MountOptions{
 		GetFilesystemOptions: vfs.GetFilesystemOptions{
-			Data: strings.Join(data, ","),
+			Data:         strings.Join(data, ","),
+			InternalData: iopts,
 		},
 		InternalMount: true,
 	}
@@ -667,3 +675,21 @@ func (c *containerMounter) makeMountPoint(ctx context.Context, creds *auth.Crede
 	}
 	return c.k.VFS().MakeSyntheticMountpoint(ctx, dest, root, creds)
 }
+
+// configureRestore returns an updated context.Context including filesystem
+// state used by restore defined by conf.
+func (c *containerMounter) configureRestore(ctx context.Context, conf *config.Config) (context.Context, error) {
+	fdmap := make(map[string]int)
+	fdmap["/"] = c.fds.remove()
+	mounts, err := c.prepareMountsVFS2()
+	if err != nil {
+		return ctx, err
+	}
+	for i := range c.mounts {
+		submount := &mounts[i]
+		if submount.fd >= 0 {
+			fdmap[submount.Destination] = submount.fd
+		}
+	}
+	return context.WithValue(ctx, gofer.CtxRestoreServerFDMap, fdmap), nil
+}