runsc: unmount volume mounts when destroy container.

PiperOrigin-RevId: 210579178
Change-Id: Iae20639c5186b1a976cbff6d05bda134cd00d0da

2 files changed, 37 insertions, 4 deletions

diff --git a/runsc/container/container.go b/runsc/container/container.go
index 16af66d3e..725b4d347 100644
--- a/runsc/container/container.go
+++ b/runsc/container/container.go
@@ -510,10 +510,6 @@ func (c *Container) Destroy() error {
 		executeHooksBestEffort(c.Spec.Hooks.Poststop, c.State())
 	}
 
-	if err := os.RemoveAll(c.Root); err != nil {
-		log.Warningf("Failed to delete container root directory %q, err: %v", c.Root, err)
-	}
-
 	// If we are the first container in the sandbox, take the sandbox down
 	// as well.
 	if c.Sandbox != nil && c.Sandbox.IsRootContainer(c.ID) {
@@ -532,6 +528,14 @@ func (c *Container) Destroy() error {
 		}
 	}
 
+	if err := destroyFS(c.Spec); err != nil {
+		return fmt.Errorf("error destroying container fs: %v", err)
+	}
+
+	if err := os.RemoveAll(c.Root); err != nil && !os.IsNotExist(err) {
+		return fmt.Errorf("error deleting container root directory %q: %v", c.Root, err)
+	}
+
 	c.Status = Stopped
 	return nil
 }
diff --git a/runsc/container/fs.go b/runsc/container/fs.go
index c12f5c331..dd8bdf120 100644
--- a/runsc/container/fs.go
+++ b/runsc/container/fs.go
@@ -134,6 +134,35 @@ func setupFS(spec *specs.Spec, conf *boot.Config, bundleDir string) error {
 	return nil
 }
 
+// destroyFS unmounts mounts done by runsc under `spec.Root.Path`. This
+// recovers the container rootfs into the original state.
+func destroyFS(spec *specs.Spec) error {
+	for _, m := range spec.Mounts {
+		if m.Type != "bind" || !specutils.IsSupportedDevMount(m) {
+			continue
+		}
+
+		// It's possible that 'm.Destination' follows symlinks inside the
+		// container.
+		dst, err := resolveSymlinks(spec.Root.Path, m.Destination)
+		if err != nil {
+			return err
+		}
+
+		flags := syscall.MNT_DETACH
+		log.Infof("Unmounting dst: %q, flags: %#x", dst, flags)
+		// Do not return error if dst is not a mountpoint.
+		// Based on http://man7.org/linux/man-pages/man2/umount.2.html
+		// For kernel version 2.6+ and MNT_DETACH flag, EINVAL means
+		// the dst is not a mount point.
+		if err := syscall.Unmount(dst, flags); err != nil &&
+			!os.IsNotExist(err) && err != syscall.EINVAL {
+			return err
+		}
+	}
+	return nil
+}
+
 // resolveSymlinks walks 'rel' having 'root' as the root directory. If there are
 // symlinks, they are evaluated relative to 'root' to ensure the end result is
 // the same as if the process was running inside the container.