Introduce FileSystem.Release()

This method gets called when a writable file gets closed in the
filesystem.  This allows filesystems to react to changed file sizes
due to writes.

fuse/api.go

@@ -43,6 +43,9 @@ type FileSystem interface {
 	// File handling
 	Open(name string, flags uint32) (file File, code Status)
 	Create(name string, flags uint32, mode uint32) (file File, code Status)
+	// Release() gets called after File.Release() on a file opened
+	// as writable.
+	Release(name string)
 	
 	// Directory handling
 	OpenDir(name string) (stream chan DirEntry, code Status)
@@ -128,6 +131,7 @@ type RawFileSystem interface {
 	Create(header *InHeader, input *CreateIn, name string) (flags uint32, handle uint64, out *EntryOut, code Status)
 	Open(header *InHeader, input *OpenIn) (flags uint32, handle uint64, status Status)
 	Read(*ReadIn, *BufferPool) ([]byte, Status)
+
 	Release(header *InHeader, input *ReleaseIn)
 	Write(*WriteIn, []byte) (written uint32, code Status)
 	Flush(*FlushIn) Status

fuse/default.go

@@ -247,6 +247,9 @@ func (me *DefaultFileSystem) Open(name string, flags uint32) (file File, code St
 	return nil, ENOSYS
 }
 
+func (me *DefaultFileSystem) Release(name string) {
+}
+
 func (me *DefaultFileSystem) OpenDir(name string) (stream chan DirEntry, status Status) {
 	return nil, ENOSYS
 }

fuse/pathfilesystem.go

@@ -213,7 +213,7 @@ func (me *FileSystemConnector) Statistics() string {
 		len(me.openFiles), len(me.inodeMap))
 }
 
-func (me *FileSystemConnector) unregisterFile(node *inode, handle uint64) interface{} {
+func (me *FileSystemConnector) unregisterFile(node *inode, handle uint64) (interface{}) {
 	me.fileLock.Lock()
 	defer me.fileLock.Unlock()
 	b, ok := me.openFiles[handle]

fuse/pathops.go

@@ -376,6 +376,22 @@ func (me *FileSystemConnector) Release(header *InHeader, input *ReleaseIn) {
 	node := me.getInodeData(header.NodeId)
 	f := me.unregisterFile(node, input.Fh).(File)
 	f.Release()
+
+	if input.Flags & O_ANYWRITE != 0 {
+		// We only signal releases to the FS if the
+		// open could have changed things.
+		var path string
+		var mount *mountData
+		me.treeLock.RLock()
+		if node.Parent != nil {
+			path, mount = node.GetPath()
+		}
+		me.treeLock.RUnlock()
+
+		if mount != nil {
+			mount.fs.Release(path)
+		}
+	}
 	me.considerDropInode(node)
 }