import "github.com/hanwen/go-fuse/fuse/nodefs"
The nodefs package offers a high level API that resembles the kernel's idea of what an FS looks like. File systems can have multiple hard-links to one file, for example. It is also suited if the data to represent fits in memory: you can construct the complete file system tree at mount time
api.go defaultfile.go defaultnode.go dir.go files.go files_linux.go fsconnector.go fsmount.go fsops.go fuse.go handle.go inode.go lockingfile.go memnode.go syscall_linux.go
const (
F_OFD_GETLK = 36
F_OFD_SETLK = 37
F_OFD_SETLKW = 38
)type File interface {
// Called upon registering the filehandle in the inode. This
// is useful in that PathFS API, where Create/Open have no
// access to the Inode at hand.
SetInode(*Inode)
// The String method is for debug printing.
String() string
// Wrappers around other File implementations, should return
// the inner file here.
InnerFile() File
Read(dest []byte, off int64) (fuse.ReadResult, fuse.Status)
Write(data []byte, off int64) (written uint32, code fuse.Status)
// File locking
GetLk(owner uint64, lk *fuse.FileLock, flags uint32, out *fuse.FileLock) (code fuse.Status)
SetLk(owner uint64, lk *fuse.FileLock, flags uint32) (code fuse.Status)
SetLkw(owner uint64, lk *fuse.FileLock, flags uint32) (code fuse.Status)
// Flush is called for close() call on a file descriptor. In
// case of duplicated descriptor, it may be called more than
// once for a file.
Flush() fuse.Status
// This is called to before the file handle is forgotten. This
// method has no return value, so nothing can synchronizes on
// the call. Any cleanup that requires specific synchronization or
// could fail with I/O errors should happen in Flush instead.
Release()
Fsync(flags int) (code fuse.Status)
// The methods below may be called on closed files, due to
// concurrency. In that case, you should return EBADF.
Truncate(size uint64) fuse.Status
GetAttr(out *fuse.Attr) fuse.Status
Chown(uid uint32, gid uint32) fuse.Status
Chmod(perms uint32) fuse.Status
Utimens(atime *time.Time, mtime *time.Time) fuse.Status
Allocate(off uint64, size uint64, mode uint32) (code fuse.Status)
}A File object is returned from FileSystem.Open and FileSystem.Create. Include the NewDefaultFile return value into the struct to inherit a null implementation.
NewDefaultFile returns a File instance that returns ENOSYS for every operation.
NewDevNullFile returns a file that accepts any write, and always returns EOF for reads.
NewLockingFile serializes operations an existing File.
LoopbackFile delegates all operations back to an underlying os.File.
NewReadOnlyFile wraps a File so all write operations are denied.
type FileSystemConnector struct {
// contains filtered or unexported fields
}FileSystemConnector translates the raw FUSE protocol (serialized structs of uint32/uint64) to operations on Go objects representing files and directories.
func Mount(mountpoint string, root Node, mountOptions *fuse.MountOptions, nodefsOptions *Options) (*fuse.Server, *FileSystemConnector, error)
Mount mounts a filesystem with the given root node on the given directory. Convenience wrapper around fuse.NewServer
func MountRoot(mountpoint string, root Node, opts *Options) (*fuse.Server, *FileSystemConnector, error)
MountRoot is like Mount but uses default fuse mount options.
func NewFileSystemConnector(root Node, opts *Options) (c *FileSystemConnector)
NewFileSystemConnector creates a FileSystemConnector with the given options.
DeleteNotify signals to the kernel that the named entry in dir for the child disappeared. No filesystem related locks should be held when calling this.
EntryNotify makes the kernel forget the entry data from the given name from a directory. After this call, the kernel will issue a new lookup request for the given name when necessary. No filesystem related locks should be held when calling this.
FileNotify notifies the kernel that data and metadata of this inode has changed. After this call completes, the kernel will issue a new GetAttr requests for metadata and new Read calls for content. Use negative offset for metadata-only invalidation, and zero-length for invalidating all content.
func (c *FileSystemConnector) FileNotifyStoreCache(node *Inode, off int64, data []byte) fuse.Status
FileNotifyStoreCache notifies the kernel about changed data of the inode.
This call is similar to FileNotify, but instead of only invalidating a data region, it puts updated data directly to the kernel cache:
After this call completes, the kernel has put updated data into the inode's cache, and will use data from that cache for non direct-IO reads from the inode in corresponding data region. After kernel's cache data is evicted, the kernel will have to issue new Read calls on user request to get data content.
ENOENT is returned if the kernel does not currently have entry for this inode in its dentry cache.
func (c *FileSystemConnector) FileRetrieveCache(node *Inode, off int64, dest []byte) (n int, st fuse.Status)
FileRetrieveCache retrieves data from kernel's inode cache.
This call retrieves data from kernel's inode cache @ offset and up to len(dest) bytes. If kernel cache has fewer consecutive data starting at offset, that fewer amount is returned. In particular if inode data at offset is not cached (0, OK) is returned.
If the kernel does not currently have entry for this inode in its dentry cache (0, OK) is still returned, pretending that the inode could be known to the kernel, but kernel's inode cache is empty.
func (c *FileSystemConnector) InodeHandleCount() int
InodeCount returns the number of inodes registered with the kernel.
func (c *FileSystemConnector) LookupNode(parent *Inode, path string) *Inode
Follows the path from the given parent, doing lookups as necessary. The path should be '/' separated without leading slash.
func (c *FileSystemConnector) Mount(parent *Inode, name string, root Node, opts *Options) fuse.Status
Mount() generates a synthetic directory node, and mounts the file system there. If opts is nil, the mount options of the root file system are inherited. The encompassing filesystem should pretend the mount point does not exist.
It returns ENOENT if the directory containing the mount point does not exist, and EBUSY if the intended mount point already exists.
Finds a node within the currently known inodes, returns the last known node and the remaining unknown path components. If parent is nil, start from FUSE mountpoint.
func (c *FileSystemConnector) RawFS() fuse.RawFileSystem
Returns the RawFileSystem so it can be mounted.
func (c *FileSystemConnector) Server() *fuse.Server
Server returns the fuse.Server that talking to the kernel.
func (c *FileSystemConnector) SetDebug(debug bool)
SetDebug toggles printing of debug information. This function is deprecated. Set the Debug option in the Options struct instead.
func (c *FileSystemConnector) Unmount(node *Inode) fuse.Status
Unmount() tries to unmount the given inode. It returns EINVAL if the path does not exist, or is not a mount point, and EBUSY if there are open files or submounts below this node.
type Inode struct {
// contains filtered or unexported fields
}An Inode reflects the kernel's idea of the inode. Inodes have IDs that are communicated to the kernel, and they have a tree structure: a directory Inode may contain named children. Each Inode object is paired with a Node object, which file system implementers should supply.
AddChild adds a child inode. The parent inode must be a directory node.
Returns any open file, preferably a r/w one.
Children returns all children of this inode.
Files() returns an opens file that have bits in common with the give mask. Use mask==0 to return all files.
FsChildren returns all the children from the same filesystem. It will skip mountpoints.
GetChild returns a child inode with the given name, or nil if it does not exist.
IsDir returns true if this is a directory.
NewChild adds a new child inode to this inode.
Node returns the file-system specific node.
Parent returns a random parent and the name this inode has under this parent. This function can be used to walk up the directory tree. It will not cross sub-mounts.
RmChild removes an inode by name, and returns it. It returns nil if child does not exist.
Print the inode. The default print method may not be used for debugging, as dumping the map requires synchronization.
type Node interface {
// Inode and SetInode are basic getter/setters. They are
// called by the FileSystemConnector. You get them for free by
// embedding the result of NewDefaultNode() in your node
// struct.
Inode() *Inode
SetInode(node *Inode)
// OnMount is called on the root node just after a mount is
// executed, either when the actual root is mounted, or when a
// filesystem is mounted in-process. The passed-in
// FileSystemConnector gives access to Notify methods and
// Debug settings.
OnMount(conn *FileSystemConnector)
// OnUnmount is executed just before a submount is removed,
// and when the process receives a forget for the FUSE root
// node.
OnUnmount()
// Lookup finds a child node to this node; it is only called
// for directory Nodes. Lookup may be called on nodes that are
// already known.
Lookup(out *fuse.Attr, name string, context *fuse.Context) (*Inode, fuse.Status)
// Deletable() should return true if this node may be discarded once
// the kernel forgets its reference.
// If it returns false, OnForget will never get called for this node. This
// is appropriate if the filesystem has no persistent backing store
// (in-memory filesystems) where discarding the node loses the stored data.
// Deletable will be called from within the treeLock critical section, so you
// cannot look at other nodes.
Deletable() bool
// OnForget is called when the kernel forgets its reference to this node and
// sends a FORGET request. It should perform cleanup and free memory as
// appropriate for the filesystem.
// OnForget is not called if the node is a directory and has children.
// This is called from within a treeLock critical section.
OnForget()
// Misc.
Access(mode uint32, context *fuse.Context) (code fuse.Status)
Readlink(c *fuse.Context) ([]byte, fuse.Status)
// Mknod should create the node, add it to the receiver's
// inode, and return it
Mknod(name string, mode uint32, dev uint32, context *fuse.Context) (newNode *Inode, code fuse.Status)
// Mkdir should create the directory Inode, add it to the
// receiver's Inode, and return it
Mkdir(name string, mode uint32, context *fuse.Context) (newNode *Inode, code fuse.Status)
Unlink(name string, context *fuse.Context) (code fuse.Status)
Rmdir(name string, context *fuse.Context) (code fuse.Status)
// Symlink should create a child inode to the receiver, and
// return it.
Symlink(name string, content string, context *fuse.Context) (*Inode, fuse.Status)
Rename(oldName string, newParent Node, newName string, context *fuse.Context) (code fuse.Status)
// Link should return the Inode of the resulting link. In
// a POSIX conformant file system, this should add 'existing'
// to the receiver, and return the Inode corresponding to
// 'existing'.
Link(name string, existing Node, context *fuse.Context) (newNode *Inode, code fuse.Status)
// Create should return an open file, and the Inode for that file.
Create(name string, flags uint32, mode uint32, context *fuse.Context) (file File, child *Inode, code fuse.Status)
// Open opens a file, and returns a File which is associated
// with a file handle. It is OK to return (nil, OK) here. In
// that case, the Node should implement Read or Write
// directly.
Open(flags uint32, context *fuse.Context) (file File, code fuse.Status)
OpenDir(context *fuse.Context) ([]fuse.DirEntry, fuse.Status)
Read(file File, dest []byte, off int64, context *fuse.Context) (fuse.ReadResult, fuse.Status)
Write(file File, data []byte, off int64, context *fuse.Context) (written uint32, code fuse.Status)
// XAttrs
GetXAttr(attribute string, context *fuse.Context) (data []byte, code fuse.Status)
RemoveXAttr(attr string, context *fuse.Context) fuse.Status
SetXAttr(attr string, data []byte, flags int, context *fuse.Context) fuse.Status
ListXAttr(context *fuse.Context) (attrs []string, code fuse.Status)
// File locking
//
// GetLk returns existing lock information for file.
GetLk(file File, owner uint64, lk *fuse.FileLock, flags uint32, out *fuse.FileLock, context *fuse.Context) (code fuse.Status)
// Sets or clears the lock described by lk on file.
SetLk(file File, owner uint64, lk *fuse.FileLock, flags uint32, context *fuse.Context) (code fuse.Status)
// Sets or clears the lock described by lk. This call blocks until the operation can be completed.
SetLkw(file File, owner uint64, lk *fuse.FileLock, flags uint32, context *fuse.Context) (code fuse.Status)
// Attributes
GetAttr(out *fuse.Attr, file File, context *fuse.Context) (code fuse.Status)
Chmod(file File, perms uint32, context *fuse.Context) (code fuse.Status)
Chown(file File, uid uint32, gid uint32, context *fuse.Context) (code fuse.Status)
Truncate(file File, size uint64, context *fuse.Context) (code fuse.Status)
Utimens(file File, atime *time.Time, mtime *time.Time, context *fuse.Context) (code fuse.Status)
Fallocate(file File, off uint64, size uint64, mode uint32, context *fuse.Context) (code fuse.Status)
StatFs() *fuse.StatfsOut
}The Node interface implements the user-defined file system functionality
NewDefaultNode returns an implementation of Node that returns ENOSYS for all operations.
NewMemNodeFSRoot creates an in-memory node-based filesystem. Files are written into a backing store under the given prefix.
type Options struct {
EntryTimeout time.Duration
AttrTimeout time.Duration
NegativeTimeout time.Duration
// If set, replace all uids with given UID.
// NewOptions() will set this to the daemon's
// uid/gid.
*fuse.Owner
// This option exists for compatibility and is ignored.
PortableInodes bool
// If set, print debug information.
Debug bool
// If set, issue Lookup rather than GetAttr calls for known
// children. This allows the filesystem to update its inode
// hierarchy in response to kernel calls.
LookupKnownChildren bool
}Options contains time out options for a node FileSystem. The default copied from libfuse and set in NewMountOptions() is (1s,1s,0s).
NewOptions generates FUSE options that correspond to libfuse's defaults.
type TreeWatcher interface {
OnAdd(parent *Inode, name string)
OnRemove(parent *Inode, name string)
}TreeWatcher is an additional interface that Nodes can implement. If they do, the OnAdd and OnRemove are called for operations on the file system tree. These functions run under a lock, so they should not do blocking operations.
type WithFlags struct {
File
// For debugging.
Description string
// Put FOPEN_* flags here.
FuseFlags uint32
// O_RDWR, O_TRUNCATE, etc.
OpenFlags uint32
}Wrap a File return in this to set FUSE flags. Also used internally to store open file data.
Package nodefs imports 10 packages (graph) and is imported by 193 packages. Updated 2019-04-26. Refresh now. Tools for package owners.