Merge pull request kata-containers#1553 from bergwolf/ro-volumes

runtime: fix virtiofsd RO volume sharing

src/runtime/virtcontainers/container.go

@@ -472,18 +472,36 @@ func (c *Container) shareFiles(ctx context.Context, m Mount, idx int, hostShared
 	} else {
 		// These mounts are created in the shared dir
 		mountDest := filepath.Join(hostMountDir, filename)
-		if err := bindMount(ctx, m.Source, mountDest, m.ReadOnly, "private"); err != nil {
-			return "", false, err
-		}
-		// Save HostPath mount value into the mount list of the container.
-		c.mounts[idx].HostPath = mountDest
-		// bindmount remount event is not propagated to mount subtrees, so we have to remount the shared dir mountpoint directly.
-		if m.ReadOnly {
-			mountDest = filepath.Join(hostSharedDir, filename)
-			if err := remountRo(c.ctx, mountDest); err != nil {
+		if !m.ReadOnly {
+			if err := bindMount(c.ctx, m.Source, mountDest, false, "private"); err != nil {
 				return "", false, err
 			}
+		} else {
+			// For RO mounts, bindmount remount event is not propagated to mount subtrees,
+			// and it doesn't present in the virtiofsd standalone mount namespace either.
+			// So we end up a bit tricky:
+			// 1. make a private bind mount to the mount source
+			// 2. make another ro bind mount on the private mount
+			// 3. move the ro bind mount to mountDest
+			// 4. umount the private bind mount created in step 1
+			privateDest := filepath.Join(getPrivatePath(c.sandboxID), filename)
+			if err := bindMount(c.ctx, m.Source, privateDest, false, "private"); err != nil {
+				return "", false, err
+			}
+			defer func() {
+				syscall.Unmount(privateDest, syscall.MNT_DETACH|UmountNoFollow)
+			}()
+			if err := bindMount(c.ctx, privateDest, privateDest, true, "private"); err != nil {
+				return "", false, err
+			}
+			if err := moveMount(c.ctx, privateDest, mountDest); err != nil {
+				return "", false, err
+			}
+
+			syscall.Unmount(privateDest, syscall.MNT_DETACH|UmountNoFollow)
 		}
+		// Save HostPath mount value into the mount list of the container.
+		c.mounts[idx].HostPath = mountDest
 	}
 
 	return guestDest, false, nil

src/runtime/virtcontainers/kata_agent.go

@@ -152,9 +152,10 @@ var kataHostSharedDir = func() string {
 }
 
 // Shared path handling:
-// 1. create two directories for each sandbox:
+// 1. create three directories for each sandbox:
 // -. /run/kata-containers/shared/sandboxes/$sbx_id/mounts/, a directory to hold all host/guest shared mounts
 // -. /run/kata-containers/shared/sandboxes/$sbx_id/shared/, a host/guest shared directory (9pfs/virtiofs source dir)
+// -. /run/kata-containers/shared/sandboxes/$sbx_id/private/, a directory to hold all temporary private mounts when creating ro mounts
 //
 // 2. /run/kata-containers/shared/sandboxes/$sbx_id/mounts/ is bind mounted readonly to /run/kata-containers/shared/sandboxes/$sbx_id/shared/, so guest cannot modify it
 //
@@ -167,6 +168,10 @@ func getMountPath(id string) string {
 	return filepath.Join(kataHostSharedDir(), id, "mounts")
 }
 
+func getPrivatePath(id string) string {
+	return filepath.Join(kataHostSharedDir(), id, "private")
+}
+
 func getSandboxPath(id string) string {
 	return filepath.Join(kataHostSharedDir(), id)
 }

src/runtime/virtcontainers/mount.go

@@ -212,6 +212,42 @@ func isDeviceMapper(major, minor int) (bool, error) {
 
 const mountPerm = os.FileMode(0755)
 
+func evalMountPath(source, destination string) (string, string, error) {
+	if source == "" {
+		return "", "", fmt.Errorf("source must be specified")
+	}
+	if destination == "" {
+		return "", "", fmt.Errorf("destination must be specified")
+	}
+
+	absSource, err := filepath.EvalSymlinks(source)
+	if err != nil {
+		return "", "", fmt.Errorf("Could not resolve symlink for source %v", source)
+	}
+
+	if err := ensureDestinationExists(absSource, destination); err != nil {
+		return "", "", fmt.Errorf("Could not create destination mount point %v: %v", destination, err)
+	}
+
+	return absSource, destination, nil
+}
+
+// moveMount moves a mountpoint to another path with some bookkeeping:
+// * evaluate all symlinks
+// * ensure the source exists
+// * recursively create the destination
+func moveMount(ctx context.Context, source, destination string) error {
+	span, _ := trace(ctx, "moveMount")
+	defer span.End()
+
+	source, destination, err := evalMountPath(source, destination)
+	if err != nil {
+		return err
+	}
+
+	return syscall.Mount(source, destination, "move", syscall.MS_MOVE, "")
+}
+
 // bindMount bind mounts a source in to a destination. This will
 // do some bookkeeping:
 // * evaluate all symlinks
@@ -222,20 +258,9 @@ func bindMount(ctx context.Context, source, destination string, readonly bool, p
 	span, _ := trace(ctx, "bindMount")
 	defer span.End()
 
-	if source == "" {
-		return fmt.Errorf("source must be specified")
-	}
-	if destination == "" {
-		return fmt.Errorf("destination must be specified")
-	}
-
-	absSource, err := filepath.EvalSymlinks(source)
+	absSource, destination, err := evalMountPath(source, destination)
 	if err != nil {
-		return fmt.Errorf("Could not resolve symlink for source %v", source)
-	}
-
-	if err := ensureDestinationExists(absSource, destination); err != nil {
-		return fmt.Errorf("Could not create destination mount point %v: %v", destination, err)
+		return err
 	}
 
 	if err := syscall.Mount(absSource, destination, "bind", syscall.MS_BIND, ""); err != nil {