Version: v2.4.0 Latest Latest

This package is not in the latest version of its module.

Go to latest
Published: Aug 28, 2023 License: BSD-3-Clause Imports: 20 Imported by: 151



Package fuse provides APIs to implement filesystems in userspace in terms of raw FUSE protocol.

A filesystem is implemented by implementing its server that provides a RawFileSystem interface. Typically the server embeds NewDefaultRawFileSystem() and implements only subset of filesystem methods:

type MyFS struct {

func NewMyFS() *MyFS {
	return &MyFS{
		RawFileSystem: fuse.NewDefaultRawFileSystem(),

// Mkdir implements "mkdir" request handler.
// For other requests - not explicitly implemented by MyFS - ENOSYS
// will be typically returned to client.
func (fs *MyFS) Mkdir(...) {

Then the filesystem can be mounted and served to a client (typically OS kernel) by creating Server:

fs := NewMyFS() // implements RawFileSystem
fssrv, err := fuse.NewServer(fs, mountpoint, &fuse.MountOptions{...})
if err != nil {

and letting the server do its work:

// either synchronously - .Serve() blocks until the filesystem is unmounted.

// or in the background - .Serve() is spawned in another goroutine, but
// before interacting with fssrv from current context we have to wait
// until the filesystem mounting is complete.
go fssrv.Serve()
err = fssrv.WaitMount()
if err != nil {

The server will serve clients by dispatching their requests to the filesystem implementation and conveying responses back. For example "mkdir" FUSE request dispatches to call

fs.Mkdir(*MkdirIn, ..., *EntryOut)

"stat" to call

fs.GetAttr(*GetAttrIn, *AttrOut)

etc. Please refer to RawFileSystem documentation for details.

Typically, each call of the API happens in its own goroutine, so take care to make the file system thread-safe.

Be careful when you access the FUSE mount from the same process. An access can tie up two OS threads (one on the request side and one on the FUSE server side). This can deadlock if there is no free thread to handle the FUSE server side. Run your program with GOMAXPROCS=1 to make the problem easier to reproduce, see https://github.com/hanwen/go-fuse/issues/261 for an example of that problem.

Higher level interfaces

As said above this packages provides way to implement filesystems in terms of raw FUSE protocol.

Package github.com/hanwen/go-fuse/v2/fs provides way to implement filesystems in terms of paths and/or inodes.

Mount styles

The NewServer() handles mounting the filesystem, which involves opening `/dev/fuse` and calling the `mount(2)` syscall. The latter needs root permissions. This is handled in one of three ways:

1) go-fuse opens `/dev/fuse` and executes the `fusermount` setuid-root helper to call `mount(2)` for us. This is the default. Does not need root permissions but needs `fusermount` installed.

2) If `MountOptions.DirectMount` is set, go-fuse calls `mount(2)` itself. Needs root permissions, but works without `fusermount`.

3) If `mountPoint` has the magic `/dev/fd/N` syntax, it means that that a privileged parent process:

* Opened /dev/fuse

* Called mount(2) on a real mountpoint directory that we don't know about

* Inherited the fd to /dev/fuse to us

* Informs us about the fd number via /dev/fd/N

This magic syntax originates from libfuse [1] and allows the FUSE server to run without any privileges and without needing `fusermount`, as the parent process performs all privileged operations.

The "privileged parent" is usually a container manager like Singularity [2], but for testing, it can also be the `mount.fuse3` helper with the `drop_privileges,setuid=$USER` flags. Example below for gocryptfs:

$ sudo mount.fuse3 "/usr/local/bin/gocryptfs#/tmp/cipher" /tmp/mnt -o drop_privileges,setuid=$USER

[1] https://github.com/libfuse/libfuse/commit/64e11073b9347fcf9c6d1eea143763ba9e946f70

[2] https://sylabs.io/guides/3.7/user-guide/bind_paths_and_mounts.html#fuse-mounts



View Source
const (

	FUSE_UNKNOWN_INO = 0xffffffff


	FUSE_LK_FLOCK = (1 << 0)

	FUSE_RELEASE_FLUSH        = (1 << 0)




	S_IFDIR = syscall.S_IFDIR
	S_IFREG = syscall.S_IFREG
	S_IFLNK = syscall.S_IFLNK
	S_IFIFO = syscall.S_IFIFO

	CUSE_INIT = 4096

	O_ANYWRITE = uint32(os.O_WRONLY | os.O_RDWR | os.O_APPEND | os.O_CREATE | os.O_TRUNC)
View Source
const (
	OK = Status(0)

	// EACCESS Permission denied
	EACCES = Status(syscall.EACCES)

	// EBUSY Device or resource busy
	EBUSY = Status(syscall.EBUSY)

	// EAGAIN Resource temporarily unavailable
	EAGAIN = Status(syscall.EAGAIN)

	// EINTR Call was interrupted
	EINTR = Status(syscall.EINTR)

	// EINVAL Invalid argument
	EINVAL = Status(syscall.EINVAL)

	// EIO I/O error
	EIO = Status(syscall.EIO)

	// ENOENT No such file or directory
	ENOENT = Status(syscall.ENOENT)

	// ENOSYS Function not implemented
	ENOSYS = Status(syscall.ENOSYS)

	// ENODATA No data available
	ENODATA = Status(syscall.ENODATA)

	// ENOTDIR Not a directory
	ENOTDIR = Status(syscall.ENOTDIR)

	// ENOTSUP Not supported
	ENOTSUP = Status(syscall.ENOTSUP)

	// EISDIR Is a directory
	EISDIR = Status(syscall.EISDIR)

	// EPERM Operation not permitted
	EPERM = Status(syscall.EPERM)

	// ERANGE Math result not representable
	ERANGE = Status(syscall.ERANGE)

	// EXDEV Cross-device link
	EXDEV = Status(syscall.EXDEV)

	// EBADF Bad file number
	EBADF = Status(syscall.EBADF)

	// ENODEV No such device
	ENODEV = Status(syscall.ENODEV)

	// EROFS Read-only file system
	EROFS = Status(syscall.EROFS)
View Source
const (
	FATTR_MODE      = (1 << 0)
	FATTR_UID       = (1 << 1)
	FATTR_GID       = (1 << 2)
	FATTR_SIZE      = (1 << 3)
	FATTR_ATIME     = (1 << 4)
	FATTR_MTIME     = (1 << 5)
	FATTR_FH        = (1 << 6)
	FATTR_ATIME_NOW = (1 << 7)
	FATTR_MTIME_NOW = (1 << 8)
	FATTR_CTIME     = (1 << 10)
View Source
const (
	// OpenOut.Flags
	FOPEN_DIRECT_IO   = (1 << 0)
	FOPEN_KEEP_CACHE  = (1 << 1)
	FOPEN_CACHE_DIR   = (1 << 3)
	FOPEN_STREAM      = (1 << 4)
View Source
const (
	CAP_ASYNC_READ          = (1 << 0)
	CAP_POSIX_LOCKS         = (1 << 1)
	CAP_FILE_OPS            = (1 << 2)
	CAP_ATOMIC_O_TRUNC      = (1 << 3)
	CAP_EXPORT_SUPPORT      = (1 << 4)
	CAP_BIG_WRITES          = (1 << 5)
	CAP_DONT_MASK           = (1 << 6)
	CAP_SPLICE_WRITE        = (1 << 7)
	CAP_SPLICE_MOVE         = (1 << 8)
	CAP_SPLICE_READ         = (1 << 9)
	CAP_FLOCK_LOCKS         = (1 << 10)
	CAP_IOCTL_DIR           = (1 << 11)
	CAP_AUTO_INVAL_DATA     = (1 << 12)
	CAP_READDIRPLUS         = (1 << 13)
	CAP_READDIRPLUS_AUTO    = (1 << 14)
	CAP_ASYNC_DIO           = (1 << 15)
	CAP_WRITEBACK_CACHE     = (1 << 16)
	CAP_NO_OPEN_SUPPORT     = (1 << 17)
	CAP_PARALLEL_DIROPS     = (1 << 18)
	CAP_HANDLE_KILLPRIV     = (1 << 19)
	CAP_POSIX_ACL           = (1 << 20)
	CAP_ABORT_ERROR         = (1 << 21)
	CAP_MAX_PAGES           = (1 << 22)
	CAP_CACHE_SYMLINKS      = (1 << 23)

To be set in InitIn/InitOut.Flags.

Keep in sync with either of * https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/uapi/linux/fuse.h * https://github.com/libfuse/libfuse/blob/master/include/fuse_kernel.h but NOT with * https://github.com/libfuse/libfuse/blob/master/include/fuse_common.h This file has CAP_HANDLE_KILLPRIV and CAP_POSIX_ACL reversed!

View Source
const (
	FUSE_IOCTL_COMPAT       = (1 << 0)
	FUSE_IOCTL_RETRY        = (1 << 2)
View Source
const (
	X_OK = 1
	W_OK = 2
	R_OK = 4
	F_OK = 0

For AccessIn.Mask.

View Source
const (
	//	NOTIFY_POLL         = -1 // notify kernel that a poll waiting for IO on a file handle should wake up
	NOTIFY_INVAL_INODE    = -2 // notify kernel that an inode should be invalidated
	NOTIFY_INVAL_ENTRY    = -3 // notify kernel that a directory entry should be invalidated
	NOTIFY_STORE_CACHE    = -4 // store data into kernel cache of an inode
	NOTIFY_RETRIEVE_CACHE = -5 // retrieve data from kernel cache of an inode
	NOTIFY_DELETE         = -6 // notify kernel that a directory entry has been deleted

View Source
const (
	WRITE_CACHE     = (1 << 0)
View Source
const (
	ENOATTR = Status(syscall.ENODATA) // On Linux, ENOATTR is an alias for ENODATA.

	// EREMOTEIO Remote I/O error
	EREMOTEIO = Status(syscall.EREMOTEIO)
View Source
const (
	// Mask for GetAttrIn.Flags. If set, GetAttrIn has a file handle set.
	FUSE_GETATTR_FH = (1 << 0)
View Source
const (
	// Linux v4.20+ caps requests at 1 MiB. Older kernels at 128 kiB.
	MAX_KERNEL_WRITE = 1024 * 1024
View Source
const (
	READ_LOCKOWNER = (1 << 1)
View Source
const RELEASE_FLUSH = (1 << 0)


This section is empty.


func NewContext

func NewContext(ctx context.Context, caller *Caller) context.Context

func Print

func Print(obj interface{}) string

Print pretty prints FUSE data types for kernel communication

func ToStatT

func ToStatT(f os.FileInfo) *syscall.Stat_t

func UtimeToTimespec

func UtimeToTimespec(t *time.Time) (ts syscall.Timespec)

UtimeToTimespec converts a "Time" pointer as passed to Utimens to a "Timespec" that can be passed to the utimensat syscall. A nil pointer is converted to the special UTIME_OMIT value.


type AccessIn

type AccessIn struct {
	Mask    uint32
	Padding uint32

type Attr

type Attr struct {
	Ino  uint64
	Size uint64

	// Blocks is the number of 512-byte blocks that the file occupies on disk.
	Blocks    uint64
	Atime     uint64
	Mtime     uint64
	Ctime     uint64
	Atimensec uint32
	Mtimensec uint32
	Ctimensec uint32
	Mode      uint32
	Nlink     uint32
	Rdev uint32

	// Blksize is the preferred size for file system operations.
	Blksize uint32
	Padding uint32

func ToAttr

func ToAttr(f os.FileInfo) *Attr

func (*Attr) AccessTime

func (a *Attr) AccessTime() time.Time

func (*Attr) ChangeTime

func (a *Attr) ChangeTime() time.Time

func (*Attr) FromStat

func (a *Attr) FromStat(s *syscall.Stat_t)

func (*Attr) IsBlock

func (a *Attr) IsBlock() bool

IsBlock reports whether the FileInfo describes a block special file.

func (*Attr) IsChar

func (a *Attr) IsChar() bool

IsChar reports whether the FileInfo describes a character special file.

func (*Attr) IsDir

func (a *Attr) IsDir() bool

IsDir reports whether the FileInfo describes a directory.

func (*Attr) IsFifo

func (a *Attr) IsFifo() bool

func (*Attr) IsRegular

func (a *Attr) IsRegular() bool

IsRegular reports whether the FileInfo describes a regular file.

func (*Attr) IsSocket

func (a *Attr) IsSocket() bool

IsSocket reports whether the FileInfo describes a socket.

func (a *Attr) IsSymlink() bool

IsSymlink reports whether the FileInfo describes a symbolic link.

func (*Attr) ModTime

func (a *Attr) ModTime() time.Time

func (*Attr) SetTimes

func (a *Attr) SetTimes(access *time.Time, mod *time.Time, chstatus *time.Time)

func (*Attr) String

func (a *Attr) String() string

type AttrOut

type AttrOut struct {
	AttrValid     uint64
	AttrValidNsec uint32
	Dummy         uint32

func (*AttrOut) SetTimeout

func (o *AttrOut) SetTimeout(dt time.Duration)

func (*AttrOut) Timeout

func (o *AttrOut) Timeout() time.Duration

type Caller

type Caller struct {
	Pid uint32

Caller has data on the process making the FS call.

The UID and GID are effective UID/GID, except for the ACCESS opcode, where UID and GID are the real UIDs

func FromContext

func FromContext(ctx context.Context) (*Caller, bool)

type Context

type Context struct {
	Cancel <-chan struct{}

Context passes along cancelation signal and request data (PID, GID, UID). The name of this class predates the standard "context" package from Go, but it does implement the context.Context interface.

When a FUSE request is canceled, the API routine should respond by returning the EINTR status code.

func (*Context) Deadline

func (c *Context) Deadline() (time.Time, bool)

func (*Context) Done

func (c *Context) Done() <-chan struct{}

func (*Context) Err

func (c *Context) Err() error

func (*Context) Value

func (c *Context) Value(key interface{}) interface{}

type CopyFileRangeIn

type CopyFileRangeIn struct {
	FhIn      uint64
	OffIn     uint64
	NodeIdOut uint64
	FhOut     uint64
	OffOut    uint64
	Len       uint64
	Flags     uint64

type CreateIn

type CreateIn struct {
	Flags uint32

	// Mode for the new file; already takes Umask into account.
	Mode uint32

	// Umask used for this create call.
	Umask   uint32
	Padding uint32

type CreateOut

type CreateOut struct {

type DirEntry

type DirEntry struct {
	// Mode is the file's mode. Only the high bits (eg. S_IFDIR)
	// are considered.
	Mode uint32

	// Name is the basename of the file in the directory.
	Name string

	// Ino is the inode number.
	Ino uint64

DirEntry is a type for PathFileSystem and NodeFileSystem to return directory contents in.

func (DirEntry) String

func (d DirEntry) String() string

type DirEntryList

type DirEntryList struct {
	// contains filtered or unexported fields

DirEntryList holds the return value for READDIR and READDIRPLUS opcodes.

func NewDirEntryList

func NewDirEntryList(data []byte, off uint64) *DirEntryList

NewDirEntryList creates a DirEntryList with the given data buffer and offset.

func (*DirEntryList) Add

func (l *DirEntryList) Add(prefix int, name string, inode uint64, mode uint32) bool

Add adds a direntry to the DirEntryList, returning whether it succeeded.

func (*DirEntryList) AddDirEntry

func (l *DirEntryList) AddDirEntry(e DirEntry) bool

AddDirEntry tries to add an entry, and reports whether it succeeded.

func (*DirEntryList) AddDirLookupEntry

func (l *DirEntryList) AddDirLookupEntry(e DirEntry) *EntryOut

AddDirLookupEntry is used for ReadDirPlus. If reserves and zeroizes space for an EntryOut struct and serializes a DirEntry. On success, it returns pointers to both structs. If not enough space was left, it returns two nil pointers.

The resulting READDIRPLUS output buffer looks like this in memory: 1) EntryOut{} 2) _Dirent{} 3) Name (null-terminated) 4) Padding to align to 8 bytes [repeat]

func (*DirEntryList) FixMode added in v2.0.3

func (l *DirEntryList) FixMode(mode uint32)

FixMode overrides the file mode of the last direntry that was added. This can be needed when a directory changes while READDIRPLUS is running. Only the file type bits of mode are considered, the rest is masked out.

type EntryOut

type EntryOut struct {
	NodeId         uint64
	Generation     uint64
	EntryValid     uint64
	AttrValid      uint64
	EntryValidNsec uint32
	AttrValidNsec  uint32

EntryOut holds the result of a (directory,name) lookup. It has two TTLs, one for the (directory, name) lookup itself, and one for the attributes (eg. size, mode). The entry TTL also applies if the lookup result is ENOENT ("negative entry lookup")

func (*EntryOut) AttrTimeout

func (o *EntryOut) AttrTimeout() time.Duration

func (*EntryOut) EntryTimeout

func (o *EntryOut) EntryTimeout() time.Duration

EntryTimeout returns entry timeout currently

func (*EntryOut) SetAttrTimeout

func (o *EntryOut) SetAttrTimeout(dt time.Duration)

func (*EntryOut) SetEntryTimeout

func (o *EntryOut) SetEntryTimeout(dt time.Duration)

type FallocateIn

type FallocateIn struct {
	Fh      uint64
	Offset  uint64
	Length  uint64
	Mode    uint32
	Padding uint32

type FileLock

type FileLock struct {
	Start uint64
	End   uint64
	Typ   uint32
	Pid   uint32

func (*FileLock) FromFlockT

func (lk *FileLock) FromFlockT(flockT *syscall.Flock_t)

func (*FileLock) ToFlockT

func (lk *FileLock) ToFlockT(flockT *syscall.Flock_t)

type FlushIn

type FlushIn struct {
	Fh        uint64
	Unused    uint32
	Padding   uint32
	LockOwner uint64

type ForgetIn

type ForgetIn struct {

	Nlookup uint64

type FsyncIn

type FsyncIn struct {
	Fh         uint64
	FsyncFlags uint32
	Padding    uint32

type GetAttrIn

type GetAttrIn struct {

	Flags_ uint32
	Dummy  uint32
	Fh_    uint64

func (*GetAttrIn) Fh

func (g *GetAttrIn) Fh() uint64

Fh accesses the file handle. This is a method, because OSXFuse does not have GetAttrIn flags.

func (*GetAttrIn) Flags

func (g *GetAttrIn) Flags() uint32

Flags accesses the flags. This is a method, because OSXFuse does not have GetAttrIn flags.

type GetXAttrIn

type GetXAttrIn struct {
	Size    uint32
	Padding uint32

type GetXAttrOut

type GetXAttrOut struct {
	Size    uint32
	Padding uint32

type InHeader

type InHeader struct {
	Length uint32
	Opcode uint32
	Unique uint64
	NodeId uint64
	Padding uint32

type InitIn

type InitIn struct {

	Major        uint32
	Minor        uint32
	MaxReadAhead uint32
	Flags        uint32

func (*InitIn) SupportsNotify

func (in *InitIn) SupportsNotify(notifyType int) bool

SupportsNotify returns whether a certain notification type is supported. Pass any of the NOTIFY_* types as argument.

func (*InitIn) SupportsVersion

func (in *InitIn) SupportsVersion(maj, min uint32) bool

SupportsVersion returns true if the kernel supports the given protocol version or newer.

type InitOut

type InitOut struct {
	Major               uint32
	Minor               uint32
	MaxReadAhead        uint32
	Flags               uint32
	MaxBackground       uint16
	CongestionThreshold uint16
	MaxWrite            uint32
	TimeGran            uint32
	MaxPages            uint16
	Padding             uint16
	Unused              [8]uint32

type InterruptIn

type InterruptIn struct {
	Unique uint64

type LatencyMap

type LatencyMap interface {
	Add(name string, dt time.Duration)

This type may be provided for recording latencies of each FUSE operation.

type LinkIn

type LinkIn struct {
	Oldnodeid uint64

type LkIn

type LkIn struct {
	Fh      uint64
	Owner   uint64
	Lk      FileLock
	LkFlags uint32
	Padding uint32

type LkOut

type LkOut struct {
	Lk FileLock

type LseekIn

type LseekIn struct {
	Fh      uint64
	Offset  uint64
	Whence  uint32
	Padding uint32

type LseekOut

type LseekOut struct {
	Offset uint64

type MkdirIn

type MkdirIn struct {

	// The mode for the new directory. The calling process' umask
	// is already factored into the mode.
	Mode  uint32
	Umask uint32

type MknodIn

type MknodIn struct {

	// Mode to use, including the Umask value
	Mode    uint32
	Rdev    uint32
	Umask   uint32
	Padding uint32

type MountOptions

type MountOptions struct {
	AllowOther bool

	// Options are passed as -o string to fusermount.
	Options []string

	// Default is _DEFAULT_BACKGROUND_TASKS, 12.  This numbers
	// controls the allowed number of requests that relate to
	// async I/O.  Concurrency for synchronous I/O is not limited.
	MaxBackground int

	// MaxWrite is the max size for read and write requests. If 0, use
	// go-fuse default (currently 64 kiB).
	// This number is internally capped at MAX_KERNEL_WRITE (higher values don't make
	// sense).
	// Non-direct-io reads are mostly served via kernel readahead, which is
	// additionally subject to the MaxReadAhead limit.
	// Implementation notes:
	// There's four values the Linux kernel looks at when deciding the request size:
	// * MaxWrite, passed via InitOut.MaxWrite. Limits the WRITE size.
	// * max_read, passed via a string mount option. Limits the READ size.
	//   go-fuse sets max_read equal to MaxWrite.
	//   You can see the current max_read value in /proc/self/mounts .
	// * MaxPages, passed via InitOut.MaxPages. In Linux 4.20 and later, the value
	//   can go up to 1 MiB and go-fuse calculates the MaxPages value acc.
	//   to MaxWrite, rounding up.
	//   On older kernels, the value is fixed at 128 kiB and the
	//   passed value is ignored. No request can be larger than MaxPages, so
	//   READ and WRITE are effectively capped at MaxPages.
	// * MaxReadAhead, passed via InitOut.MaxReadAhead.
	MaxWrite int

	// MaxReadAhead is the max read ahead size to use. It controls how much data the
	// kernel reads in advance to satisfy future read requests from applications.
	// How much exactly is subject to clever heuristics in the kernel
	// (see https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/readahead.c?h=v6.2-rc5#n375
	// if you are brave) and hence also depends on the kernel version.
	// If 0, use kernel default. This number is capped at the kernel maximum
	// (128 kiB on Linux) and cannot be larger than MaxWrite.
	// MaxReadAhead only affects buffered reads (=non-direct-io), but even then, the
	// kernel can and does send larger reads to satisfy read reqests from applications
	// (up to MaxWrite or VM_READAHEAD_PAGES=128 kiB, whichever is less).
	MaxReadAhead int

	// If IgnoreSecurityLabels is set, all security related xattr
	// requests will return NO_DATA without passing through the
	// user defined filesystem.  You should only set this if you
	// file system implements extended attributes, and you are not
	// interested in security labels.
	IgnoreSecurityLabels bool // ignoring labels should be provided as a fusermount mount option.

	// If RememberInodes is set, we will never forget inodes.
	// This may be useful for NFS.
	RememberInodes bool

	// Values shown in "df -T" and friends
	// First column, "Filesystem"
	FsName string

	// Second column, "Type", will be shown as "fuse." + Name
	Name string

	// If set, wrap the file system in a single-threaded locking wrapper.
	SingleThreaded bool

	// If set, return ENOSYS for Getxattr calls, so the kernel does not issue any
	// Xattr operations at all.
	DisableXAttrs bool

	// If set, print debugging information.
	Debug bool

	// If set, sink for debug statements.
	Logger *log.Logger

	// If set, ask kernel to forward file locks to FUSE. If using,
	// you must implement the GetLk/SetLk/SetLkw methods.
	EnableLocks bool

	// If set, the kernel caches all Readlink return values. The
	// filesystem must use content notification to force the
	// kernel to issue a new Readlink call.
	EnableSymlinkCaching bool

	// If set, ask kernel not to do automatic data cache invalidation.
	// The filesystem is fully responsible for invalidating data cache.
	ExplicitDataCacheControl bool

	// SyncRead is off by default, which means that go-fuse enable the
	// FUSE_CAP_ASYNC_READ capability.
	// The kernel then submits multiple concurrent reads to service
	// userspace requests and kernel readahead.
	// Setting SyncRead disables the FUSE_CAP_ASYNC_READ capability.
	// The kernel then only sends one read request per file handle at a time,
	// and orders the requests by offset.
	// This is useful if reading out of order or concurrently is expensive for
	// (example: Amazon Cloud Drive).
	// See the comment to FUSE_CAP_ASYNC_READ in
	// https://github.com/libfuse/libfuse/blob/master/include/fuse_common.h
	// for more details.
	SyncRead bool

	// If set, fuse will first attempt to use syscall.Mount instead of
	// fusermount to mount the filesystem. This will not update /etc/mtab
	// but might be needed if fusermount is not available.
	// Also, Server.Unmount will attempt syscall.Unmount before calling
	// fusermount.
	DirectMount bool

	// DirectMountStrict is like DirectMount but no fallback to fusermount is
	// performed. If both DirectMount and DirectMountStrict are set,
	// DirectMountStrict wins.
	DirectMountStrict bool

	// DirectMountFlags are the mountflags passed to syscall.Mount. If zero, the
	// default value used by fusermount are used: syscall.MS_NOSUID|syscall.MS_NODEV.
	// If you actually *want* zero flags, pass syscall.MS_MGC_VAL, which is ignored
	// by the kernel. See `man 2 mount` for details about MS_MGC_VAL.
	DirectMountFlags uintptr

	// EnableAcls enables kernel ACL support.
	// See the comments to FUSE_CAP_POSIX_ACL
	// in https://github.com/libfuse/libfuse/blob/master/include/fuse_common.h
	// for details.
	EnableAcl bool

	// Disable ReadDirPlus capability so ReadDir is used instead. Simple
	// directory queries (i.e. 'ls' without '-l') can be faster with
	// ReadDir, as no per-file stat calls are needed
	DisableReadDirPlus bool

type NotifyInvalDeleteOut

type NotifyInvalDeleteOut struct {
	Parent  uint64
	Child   uint64
	NameLen uint32
	Padding uint32

type NotifyInvalEntryOut

type NotifyInvalEntryOut struct {
	Parent  uint64
	NameLen uint32
	Padding uint32

type NotifyInvalInodeOut

type NotifyInvalInodeOut struct {
	Ino    uint64
	Off    int64
	Length int64

type NotifyRetrieveIn

type NotifyRetrieveIn struct {
	Dummy1 uint64
	Offset uint64
	Size   uint32
	Dummy2 uint32
	Dummy3 uint64
	Dummy4 uint64

type NotifyRetrieveOut

type NotifyRetrieveOut struct {
	NotifyUnique uint64
	Nodeid       uint64
	Offset       uint64
	Size         uint32
	Padding      uint32

type NotifyStoreOut

type NotifyStoreOut struct {
	Nodeid  uint64
	Offset  uint64
	Size    uint32
	Padding uint32

type OpenIn

type OpenIn struct {
	Flags uint32
	Mode  uint32

type OpenOut

type OpenOut struct {
	Fh        uint64
	OpenFlags uint32
	Padding   uint32

type OutHeader

type OutHeader struct {
	Length uint32
	Status int32
	Unique uint64

type Owner

type Owner struct {
	Uid uint32
	Gid uint32

func CurrentOwner

func CurrentOwner() *Owner

type RawFileSystem

type RawFileSystem interface {
	String() string

	// If called, provide debug output through the log package.
	SetDebug(debug bool)

	// Lookup is called by the kernel when the VFS wants to know
	// about a file inside a directory. Many lookup calls can
	// occur in parallel, but only one call happens for each (dir,
	// name) pair.
	Lookup(cancel <-chan struct{}, header *InHeader, name string, out *EntryOut) (status Status)

	// Forget is called when the kernel discards entries from its
	// dentry cache. This happens on unmount, and when the kernel
	// is short on memory. Since it is not guaranteed to occur at
	// any moment, and since there is no return value, Forget
	// should not do I/O, as there is no channel to report back
	// I/O errors.
	Forget(nodeid, nlookup uint64)

	// Attributes.
	GetAttr(cancel <-chan struct{}, input *GetAttrIn, out *AttrOut) (code Status)
	SetAttr(cancel <-chan struct{}, input *SetAttrIn, out *AttrOut) (code Status)

	// Modifying structure.
	Mknod(cancel <-chan struct{}, input *MknodIn, name string, out *EntryOut) (code Status)
	Mkdir(cancel <-chan struct{}, input *MkdirIn, name string, out *EntryOut) (code Status)
	Unlink(cancel <-chan struct{}, header *InHeader, name string) (code Status)
	Rmdir(cancel <-chan struct{}, header *InHeader, name string) (code Status)
	Rename(cancel <-chan struct{}, input *RenameIn, oldName string, newName string) (code Status)
	Link(cancel <-chan struct{}, input *LinkIn, filename string, out *EntryOut) (code Status)

	Symlink(cancel <-chan struct{}, header *InHeader, pointedTo string, linkName string, out *EntryOut) (code Status)
	Readlink(cancel <-chan struct{}, header *InHeader) (out []byte, code Status)
	Access(cancel <-chan struct{}, input *AccessIn) (code Status)

	// GetXAttr reads an extended attribute, and should return the
	// number of bytes. If the buffer is too small, return ERANGE,
	// with the required buffer size.
	GetXAttr(cancel <-chan struct{}, header *InHeader, attr string, dest []byte) (sz uint32, code Status)

	// ListXAttr lists extended attributes as '\0' delimited byte
	// slice, and return the number of bytes. If the buffer is too
	// small, return ERANGE, with the required buffer size.
	ListXAttr(cancel <-chan struct{}, header *InHeader, dest []byte) (uint32, Status)

	// SetAttr writes an extended attribute.
	SetXAttr(cancel <-chan struct{}, input *SetXAttrIn, attr string, data []byte) Status

	// RemoveXAttr removes an extended attribute.
	RemoveXAttr(cancel <-chan struct{}, header *InHeader, attr string) (code Status)

	// File handling.
	Create(cancel <-chan struct{}, input *CreateIn, name string, out *CreateOut) (code Status)
	Open(cancel <-chan struct{}, input *OpenIn, out *OpenOut) (status Status)
	Read(cancel <-chan struct{}, input *ReadIn, buf []byte) (ReadResult, Status)
	Lseek(cancel <-chan struct{}, in *LseekIn, out *LseekOut) Status

	// File locking
	GetLk(cancel <-chan struct{}, input *LkIn, out *LkOut) (code Status)
	SetLk(cancel <-chan struct{}, input *LkIn) (code Status)
	SetLkw(cancel <-chan struct{}, input *LkIn) (code Status)

	Release(cancel <-chan struct{}, input *ReleaseIn)
	Write(cancel <-chan struct{}, input *WriteIn, data []byte) (written uint32, code Status)
	CopyFileRange(cancel <-chan struct{}, input *CopyFileRangeIn) (written uint32, code Status)

	Flush(cancel <-chan struct{}, input *FlushIn) Status
	Fsync(cancel <-chan struct{}, input *FsyncIn) (code Status)
	Fallocate(cancel <-chan struct{}, input *FallocateIn) (code Status)

	// Directory handling
	OpenDir(cancel <-chan struct{}, input *OpenIn, out *OpenOut) (status Status)
	ReadDir(cancel <-chan struct{}, input *ReadIn, out *DirEntryList) Status
	ReadDirPlus(cancel <-chan struct{}, input *ReadIn, out *DirEntryList) Status
	ReleaseDir(input *ReleaseIn)
	FsyncDir(cancel <-chan struct{}, input *FsyncIn) (code Status)

	StatFs(cancel <-chan struct{}, input *InHeader, out *StatfsOut) (code Status)

	// This is called on processing the first request. The
	// filesystem implementation can use the server argument to
	// talk back to the kernel (through notify methods).

RawFileSystem is an interface close to the FUSE wire protocol.

Unless you really know what you are doing, you should not implement this, but rather the nodefs.Node or pathfs.FileSystem interfaces; the details of getting interactions with open files, renames, and threading right etc. are somewhat tricky and not very interesting.

Each FUSE request results in a corresponding method called by Server. Several calls may be made simultaneously, because the server typically calls each method in separate goroutine.

A null implementation is provided by NewDefaultRawFileSystem.

After a successful FUSE API call returns, you may not read input or write output data: for performance reasons, memory is reused for following requests, and reading/writing the request data will lead to race conditions. If you spawn a background routine from a FUSE API call, any incoming request data it wants to reference should be copied over.

If a FUSE API call is canceled (which is signaled by closing the `cancel` channel), the API call should return EINTR. In this case, the outstanding request data is not reused, so the API call may return EINTR without ensuring that child contexts have successfully completed.

func NewDefaultRawFileSystem

func NewDefaultRawFileSystem() RawFileSystem

NewDefaultRawFileSystem returns ENOSYS (not implemented) for all operations.

type ReadIn

type ReadIn struct {
	Fh        uint64
	Offset    uint64
	Size      uint32
	ReadFlags uint32
	LockOwner uint64
	Flags     uint32
	Padding   uint32

type ReadResult

type ReadResult interface {
	// Returns the raw bytes for the read, possibly using the
	// passed buffer. The buffer should be larger than the return
	// value from Size.
	Bytes(buf []byte) ([]byte, Status)

	// Size returns how many bytes this return value takes at most.
	Size() int

	// Done() is called after sending the data to the kernel.

The result of Read is an array of bytes, but for performance reasons, we can also return data as a file-descriptor/offset/size tuple. If the backing store for a file is another filesystem, this reduces the amount of copying between the kernel and the FUSE server. The ReadResult interface captures both cases.

func ReadResultData

func ReadResultData(b []byte) ReadResult

func ReadResultFd

func ReadResultFd(fd uintptr, off int64, sz int) ReadResult

type ReleaseIn

type ReleaseIn struct {
	Fh           uint64
	Flags        uint32
	ReleaseFlags uint32
	LockOwner    uint64

type Rename1In

type Rename1In struct {
	Newdir uint64

type RenameIn

type RenameIn struct {
	Newdir  uint64
	Flags   uint32
	Padding uint32

type Server

type Server struct {
	// contains filtered or unexported fields

Server contains the logic for reading from the FUSE device and translating it to RawFileSystem interface calls.

func NewServer

func NewServer(fs RawFileSystem, mountPoint string, opts *MountOptions) (*Server, error)

NewServer creates a FUSE server and attaches ("mounts") it to the `mountPoint` directory.

See the "Mount styles" section in the package documentation if you want to know about the inner workings of the mount process. Usually you do not.

func (*Server) DebugData

func (ms *Server) DebugData() string

DebugData returns internal status information for debugging purposes.

func (*Server) DeleteNotify

func (ms *Server) DeleteNotify(parent uint64, child uint64, name string) Status

DeleteNotify notifies the kernel that an entry is removed from a directory. In many cases, this is equivalent to EntryNotify, except when the directory is in use, eg. as working directory of some process. You should not hold any FUSE filesystem locks, as that can lead to deadlock.

func (*Server) EntryNotify

func (ms *Server) EntryNotify(parent uint64, name string) Status

EntryNotify should be used if the existence status of an entry within a directory changes. You should not hold any FUSE filesystem locks, as that can lead to deadlock.

func (*Server) InodeNotify

func (ms *Server) InodeNotify(node uint64, off int64, length int64) Status

InodeNotify invalidates the information associated with the inode (ie. data cache, attributes, etc.)

func (*Server) InodeNotifyStoreCache

func (ms *Server) InodeNotifyStoreCache(node uint64, offset int64, data []byte) Status

InodeNotifyStoreCache tells kernel to store data into inode's cache.

This call is similar to InodeNotify, but instead of only invalidating a data region, it gives updated data directly to the kernel.

func (*Server) InodeRetrieveCache

func (ms *Server) InodeRetrieveCache(node uint64, offset int64, dest []byte) (n int, st Status)

InodeRetrieveCache retrieves data from kernel's inode cache.

InodeRetrieveCache asks kernel to return data from its cache for inode at [offset:offset+len(dest)) and waits for corresponding reply. 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.

The kernel returns ENOENT if it does not currently have entry for this inode in its dentry cache.

func (*Server) KernelSettings

func (ms *Server) KernelSettings() *InitIn

KernelSettings returns the Init message from the kernel, so filesystems can adapt to availability of features of the kernel driver. The message should not be altered.

func (*Server) RecordLatencies

func (ms *Server) RecordLatencies(l LatencyMap)

RecordLatencies switches on collection of timing for each request coming from the kernel.P assing a nil argument switches off the

func (*Server) Serve

func (ms *Server) Serve()

Serve initiates the FUSE loop. Normally, callers should run Serve() and wait for it to exit, but tests will want to run this in a goroutine.

Each filesystem operation executes in a separate goroutine.

func (*Server) SetDebug

func (ms *Server) SetDebug(dbg bool)

SetDebug is deprecated. Use MountOptions.Debug instead.

func (*Server) Unmount

func (ms *Server) Unmount() (err error)

Unmount calls fusermount -u on the mount. This has the effect of shutting down the filesystem. After the Server is unmounted, it should be discarded. This function is idempotent.

Does not work when we were mounted with the magic /dev/fd/N mountpoint syntax, as we do not know the real mountpoint. Unmount using

fusermount -u /path/to/real/mountpoint

in this case.

func (*Server) Wait

func (ms *Server) Wait()

Wait waits for the serve loop to exit. This should only be called after Serve has been called, or it will hang indefinitely.

func (*Server) WaitMount

func (ms *Server) WaitMount() error

WaitMount waits for the first request to be served. Use this to avoid racing between accessing the (empty or not yet mounted) mountpoint, and the OS trying to setup the user-space mount.

type SetAttrIn

type SetAttrIn struct {

type SetAttrInCommon

type SetAttrInCommon struct {

	Valid     uint32
	Padding   uint32
	Fh        uint64
	Size      uint64
	LockOwner uint64
	Atime     uint64
	Mtime     uint64
	Ctime     uint64
	Atimensec uint32
	Mtimensec uint32
	Ctimensec uint32
	Mode      uint32
	Unused4   uint32
	Unused5 uint32

func (*SetAttrInCommon) GetATime

func (s *SetAttrInCommon) GetATime() (time.Time, bool)

func (*SetAttrInCommon) GetCTime

func (s *SetAttrInCommon) GetCTime() (time.Time, bool)

func (*SetAttrInCommon) GetFh

func (s *SetAttrInCommon) GetFh() (uint64, bool)

GetFh returns the file handle if available, or 0 if undefined.

func (*SetAttrInCommon) GetGID

func (s *SetAttrInCommon) GetGID() (uint32, bool)

func (*SetAttrInCommon) GetMTime

func (s *SetAttrInCommon) GetMTime() (time.Time, bool)

func (*SetAttrInCommon) GetMode

func (s *SetAttrInCommon) GetMode() (uint32, bool)

func (*SetAttrInCommon) GetSize

func (s *SetAttrInCommon) GetSize() (uint64, bool)

func (*SetAttrInCommon) GetUID

func (s *SetAttrInCommon) GetUID() (uint32, bool)

type SetXAttrIn

type SetXAttrIn struct {
	Size  uint32
	Flags uint32

type StatfsOut

type StatfsOut struct {
	Blocks  uint64
	Bfree   uint64
	Bavail  uint64
	Files   uint64
	Ffree   uint64
	Bsize   uint32
	NameLen uint32
	Frsize  uint32
	Padding uint32
	Spare   [6]uint32

func (*StatfsOut) FromStatfsT

func (s *StatfsOut) FromStatfsT(statfs *syscall.Statfs_t)

type Status

type Status int32

Status is the errno number that a FUSE call returns to the kernel.

func ToStatus

func ToStatus(err error) Status

ToStatus extracts an errno number from Go error objects. If it fails, it logs an error and returns ENOSYS.

func (Status) Ok

func (code Status) Ok() bool

func (Status) String

func (code Status) String() string

type WriteIn

type WriteIn struct {
	Fh         uint64
	Offset     uint64
	Size       uint32
	WriteFlags uint32
	LockOwner  uint64
	Flags      uint32
	Padding    uint32

type WriteOut

type WriteOut struct {
	Size    uint32
	Padding uint32


Path Synopsis
This package is deprecated.
This package is deprecated.
This package is deprecated.
This package is deprecated.
Package test holds the tests for Go-FUSE and is not for end-user consumption.
Package test holds the tests for Go-FUSE and is not for end-user consumption.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL