Add env var to toggle pivot root or ms_move

Use the  DOCKER_RAMDISK env var to tell the native driver not to use
a pivot root when setting up the rootfs of a container.
Docker-DCO-1.1-Signed-off-by: Michael Crosby <michael@crosbymichael.com> (github: crosbymichael)

libcontainer/container.go

@@ -13,6 +13,7 @@ type Context map[string]string
 type Container struct {
 	Hostname     string          `json:"hostname,omitempty"`      // hostname
 	ReadonlyFs   bool            `json:"readonly_fs,omitempty"`   // set the containers rootfs as readonly
+	NoPivotRoot  bool            `json:"no_pivot_root,omitempty"` // this can be enabled if you are running in ramdisk
 	User         string          `json:"user,omitempty"`          // user to execute the process as
 	WorkingDir   string          `json:"working_dir,omitempty"`   // current working directory
 	Env          []string        `json:"environment,omitempty"`   // environment to set

libcontainer/nsinit/init.go

@@ -51,7 +51,7 @@ func (ns *linuxNs) Init(container *libcontainer.Container, uncleanRootfs, consol
 	if err := system.ParentDeathSignal(); err != nil {
 		return fmt.Errorf("parent death signal %s", err)
 	}
-	if err := setupNewMountNamespace(rootfs, console, container.ReadonlyFs); err != nil {
+	if err := setupNewMountNamespace(rootfs, console, container.ReadonlyFs, container.NoPivotRoot); err != nil {
 		return fmt.Errorf("setup mount namespace %s", err)
 	}
 	if err := setupNetwork(container, context); err != nil {

libcontainer/nsinit/mount.go

@@ -19,9 +19,12 @@ const defaultMountFlags = syscall.MS_NOEXEC | syscall.MS_NOSUID | syscall.MS_NOD
 //
 // There is no need to unmount the new mounts because as soon as the mount namespace
 // is no longer in use, the mounts will be removed automatically
-func setupNewMountNamespace(rootfs, console string, readonly bool) error {
+func setupNewMountNamespace(rootfs, console string, readonly, noPivotRoot bool) error {
-	// mount as slave so that the new mounts do not propagate to the host
+	flag := syscall.MS_PRIVATE
-	if err := system.Mount("", "/", "", syscall.MS_PRIVATE|syscall.MS_REC, ""); err != nil {
+	if noPivotRoot {
+		flag = syscall.MS_SLAVE
+	}
+	if err := system.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil {
 		return fmt.Errorf("mounting / as slave %s", err)
 	}
 	if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil {
@@ -52,6 +55,23 @@ func setupNewMountNamespace(rootfs, console string, readonly bool) error {
 		return fmt.Errorf("chdir into %s %s", rootfs, err)
 	}
 
+	if noPivotRoot {
+		if err := rootMsMove(rootfs); err != nil {
+			return err
+		}
+	} else {
+		if err := rootPivot(rootfs); err != nil {
+			return err
+		}
+	}
+
+	system.Umask(0022)
+
+	return nil
+}
+
+// use a pivot root to setup the rootfs
+func rootPivot(rootfs string) error {
 	pivotDir, err := ioutil.TempDir(rootfs, ".pivot_root")
 	if err != nil {
 		return fmt.Errorf("can't create pivot_root dir %s", pivotDir, err)
@@ -62,20 +82,28 @@ func setupNewMountNamespace(rootfs, console string, readonly bool) error {
 	if err := system.Chdir("/"); err != nil {
 		return fmt.Errorf("chdir / %s", err)
 	}
-
 	// path to pivot dir now changed, update
 	pivotDir = filepath.Join("/", filepath.Base(pivotDir))
-
 	if err := system.Unmount(pivotDir, syscall.MNT_DETACH); err != nil {
 		return fmt.Errorf("unmount pivot_root dir %s", err)
 	}
-
 	if err := os.Remove(pivotDir); err != nil {
 		return fmt.Errorf("remove pivot_root dir %s", err)
 	}
+	return nil
+}
 
-	system.Umask(0022)
+// use MS_MOVE and chroot to setup the rootfs
-
+func rootMsMove(rootfs string) error {
+	if err := system.Mount(rootfs, "/", "", syscall.MS_MOVE, ""); err != nil {
+		return fmt.Errorf("mount move %s into / %s", rootfs, err)
+	}
+	if err := system.Chroot("."); err != nil {
+		return fmt.Errorf("chroot . %s", err)
+	}
+	if err := system.Chdir("/"); err != nil {
+		return fmt.Errorf("chdir / %s", err)
+	}
 	return nil
 }