mount: support direct-io for loopback device

Signed-off-by: Wei Fu <fuweid89@gmail.com>

mount/losetup_linux.go

@@ -89,6 +89,12 @@ func setupLoopDev(backingFile, loopDev string, param LoopParams) (_ *os.File, re
 		return nil, fmt.Errorf("could not set loop fd for device: %s: %w", loopDev, err)
 	}
 
+	defer func() {
+		if retErr != nil {
+			_ = unix.IoctlSetInt(int(loop.Fd()), unix.LOOP_CLR_FD, 0)
+		}
+	}()
+
 	// 3. Set Info
 	info := unix.LoopInfo64{}
 	copy(info.File_name[:], backingFile)
@@ -100,27 +106,20 @@ func setupLoopDev(backingFile, loopDev string, param LoopParams) (_ *os.File, re
 		info.Flags |= unix.LO_FLAGS_AUTOCLEAR
 	}
 
-	if param.Direct {
-		info.Flags |= unix.LO_FLAGS_DIRECT_IO
-	}
-
 	err = unix.IoctlLoopSetStatus64(int(loop.Fd()), &info)
-	if err == nil {
+	if err != nil {
-		return loop, nil
+		return nil, fmt.Errorf("failed to set loop device info: %w", err)
 	}
 
+	// 4. Set Direct IO
 	if param.Direct {
-		// Retry w/o direct IO flag in case kernel does not support it. The downside is that
+		err = unix.IoctlSetInt(int(loop.Fd()), unix.LOOP_SET_DIRECT_IO, 1)
-		// it will suffer from double cache problem.
+		if err != nil {
-		info.Flags &= ^(uint32(unix.LO_FLAGS_DIRECT_IO))
+			return nil, fmt.Errorf("failed to setup loop with direct: %w", err)
-		err = unix.IoctlLoopSetStatus64(int(loop.Fd()), &info)
-		if err == nil {
-			return loop, nil
 		}
 	}
 
-	_ = unix.IoctlSetInt(int(loop.Fd()), unix.LOOP_CLR_FD, 0)
+	return loop, nil
-	return nil, fmt.Errorf("failed to set loop device info: %v", err)
 }
 
 // setupLoop looks for (and possibly creates) a free loop device, and

mount/mount_linux.go

@@ -63,9 +63,6 @@ func (m *Mount) mount(target string) (err error) {
 	}
 
 	flags, data, losetup := parseMountOptions(options)
-	if len(data) > pagesize {
-		return errors.New("mount options is too long")
-	}
 
 	// propagation types.
 	const ptypes = unix.MS_SHARED | unix.MS_PRIVATE | unix.MS_SLAVE | unix.MS_UNBINDABLE
@@ -73,15 +70,27 @@ func (m *Mount) mount(target string) (err error) {
 	// Ensure propagation type change flags aren't included in other calls.
 	oflags := flags &^ ptypes
 
+	var loopParams LoopParams
+	if losetup {
+		loopParams = LoopParams{
+			Readonly:  oflags&unix.MS_RDONLY == unix.MS_RDONLY,
+			Autoclear: true,
+		}
+		loopParams.Direct, data = hasDirectIO(data)
+	}
+
+	dataInStr := strings.Join(data, ",")
+	if len(dataInStr) > pagesize {
+		return errors.New("mount options is too long")
+	}
+
 	// In the case of remounting with changed data (data != ""), need to call mount (moby/moby#34077).
-	if flags&unix.MS_REMOUNT == 0 || data != "" {
+	if flags&unix.MS_REMOUNT == 0 || dataInStr != "" {
 		// Initial call applying all non-propagation flags for mount
 		// or remount with changed data
 		source := m.Source
 		if losetup {
-			loFile, err := setupLoop(m.Source, LoopParams{
+			loFile, err := setupLoop(m.Source, loopParams)
-				Readonly:  oflags&unix.MS_RDONLY == unix.MS_RDONLY,
-				Autoclear: true})
 			if err != nil {
 				return err
 			}
@@ -90,7 +99,7 @@ func (m *Mount) mount(target string) (err error) {
 			// Mount the loop device instead
 			source = loFile.Name()
 		}
-		if err := mountAt(chdir, source, target, m.Type, uintptr(oflags), data); err != nil {
+		if err := mountAt(chdir, source, target, m.Type, uintptr(oflags), dataInStr); err != nil {
 			return err
 		}
 	}
@@ -199,7 +208,7 @@ func UnmountAll(mount string, flags int) error {
 
 // parseMountOptions takes fstab style mount options and parses them for
 // use with a standard mount() syscall
-func parseMountOptions(options []string) (int, string, bool) {
+func parseMountOptions(options []string) (int, []string, bool) {
 	var (
 		flag    int
 		losetup bool
@@ -252,7 +261,16 @@ func parseMountOptions(options []string) (int, string, bool) {
 			data = append(data, o)
 		}
 	}
-	return flag, strings.Join(data, ","), losetup
+	return flag, data, losetup
+}
+
+func hasDirectIO(opts []string) (bool, []string) {
+	for idx, opt := range opts {
+		if opt == "direct-io" {
+			return true, append(opts[:idx], opts[idx+1:]...)
+		}
+	}
+	return false, opts
 }
 
 // compactLowerdirOption updates overlay lowdir option and returns the common