Merge branch 'master' into y/nodefs-cancel

Some notable patches, nothing major:

- 36b35911 "fuse: fix deadlock with parallel mounts"

- f91e3081 "fuse: debug log: print PID"

- 334a5a5f "fuse: update for kernel API version 40"

- afb9051a "fuse: provide InitIn/InitOut.Flags64()"

There are other changes which add e.g. passthrough support and other
features, but they apply mostly to fs package, not nodefs, and I
reviewed them to be safe and orthogonal to what WCFS uses and needs.

* master: (53 commits)
  fs: skip passthrough test for older kernels
  fuse: provide InitIn/InitOut.Flags64()
  fs: reorganize code
  fs: support passthrough mode
  fuse: expose BackingFdMap
  fuse: fix printing of new constant
  fuse: remove deprecated CAP_DIRECT_IO_RELAX
  fuse: print tweaks for passthrough support
  fs: return fileEntry from registerFile and addNewChild
  benchmark: add a test reading from libfuse passthrough_hp
  benchmark: modernize setupFS
  benchmark: introduce benchmark for loopback reads
  benchmark: simplify ReadLines
  fs: fix stray comment
  fuse: wire up passthrough support
  fuse: point comment to fs API rather than nodefs/pathfs
  fuse: move Flags2 handling to platform specific code
  fuse: update for kernel API version 40
  fuse: fix CreateOut print
  fuse: document singleReader background
  ...

.github/workflows/ci.yml

@@ -11,7 +11,6 @@ jobs:
     strategy:
       matrix:
         go:
-          - "1.16.x" # Golang upstream stable
           - "1.17.x" # Golang upstream stable
           - "1.18.x" # Golang upstream stable
           - "1.19.x" # Golang upstream stable

README.md

@@ -106,6 +106,7 @@ output. Here is how to read it:
 - `tA` and `tE` means timeout for attributes and directory entry correspondingly;
 - `[<off> +<size>)` means data range from `<off>` inclusive till `<off>+<size>` exclusive;
 - `Xb` means `X bytes`.
+- `pX` means the request originated from PID `x`. 0 means the request originated from the kernel.
 
 Every line is prefixed with either `rx <unique>` or `tx <unique>` to denote
 whether it was for kernel request, which Go-FUSE received, or reply, which
@@ -114,24 +115,24 @@ Go-FUSE sent back to kernel.
 Example debug log output:
 
 ```
-rx 2: LOOKUP i1 [".wcfs"] 6b
+rx 2: LOOKUP i1 [".wcfs"] 6b p5874
 tx 2:     OK, {i3 g2 tE=1s tA=1s {M040755 SZ=0 L=0 1000:1000 B0*0 i0:3 A 0.000000 M 0.000000 C 0.000000}}
-rx 3: LOOKUP i3 ["zurl"] 5b
+rx 3: LOOKUP i3 ["zurl"] 5b p5874
 tx 3:     OK, {i4 g3 tE=1s tA=1s {M0100644 SZ=33 L=1 1000:1000 B0*0 i0:4 A 0.000000 M 0.000000 C 0.000000}}
-rx 4: OPEN i4 {O_RDONLY,0x8000}
+rx 4: OPEN i4 {O_RDONLY,0x8000} p5874
 tx 4:     38=function not implemented, {Fh 0 }
-rx 5: READ i4 {Fh 0 [0 +4096)  L 0 RDONLY,0x8000}
+rx 5: READ i4 {Fh 0 [0 +4096)  L 0 RDONLY,0x8000} p5874
 tx 5:     OK,  33b data "file:///"...
-rx 6: GETATTR i4 {Fh 0}
+rx 6: GETATTR i4 {Fh 0} p5874
 tx 6:     OK, {tA=1s {M0100644 SZ=33 L=1 1000:1000 B0*0 i0:4 A 0.000000 M 0.000000 C 0.000000}}
-rx 7: FLUSH i4 {Fh 0}
+rx 7: FLUSH i4 {Fh 0} p5874
 tx 7:     OK
-rx 8: LOOKUP i1 ["head"] 5b
+rx 8: LOOKUP i1 ["head"] 5b p5874
 tx 8:     OK, {i5 g4 tE=1s tA=1s {M040755 SZ=0 L=0 1000:1000 B0*0 i0:5 A 0.000000 M 0.000000 C 0.000000}}
-rx 9: LOOKUP i5 ["bigfile"] 8b
+rx 9: LOOKUP i5 ["bigfile"] 8b p5874
 tx 9:     OK, {i6 g5 tE=1s tA=1s {M040755 SZ=0 L=0 1000:1000 B0*0 i0:6 A 0.000000 M 0.000000 C 0.000000}}
-rx 10: FLUSH i4 {Fh 0}
+rx 10: FLUSH i4 {Fh 0} p5874
 tx 10:     OK
-rx 11: GETATTR i1 {Fh 0}
+rx 11: GETATTR i1 {Fh 0} p5874
 tx 11:     OK, {tA=1s {M040755 SZ=0 L=1 1000:1000 B0*0 i0:1 A 0.000000 M 0.000000 C 0.000000}}
 ```

all.bash

@@ -17,7 +17,7 @@ GO_TEST="go test -timeout 5m -p 1 -count 1"
 # Run all tests as current user
 $GO_TEST ./...
 # Direct-mount tests need to run as root
-sudo env PATH=$PATH $GO_TEST -run TestDirectMount ./fs ./fuse
+sudo env PATH=$PATH $GO_TEST -run 'Test(DirectMount|Passthrough)' ./fs ./fuse
 
 make -C benchmark
 go test ./benchmark -test.bench '.*' -test.cpu 1,2

benchmark/benchmark.go

@@ -7,32 +7,22 @@ package benchmark
 // Routines for benchmarking fuse.
 
 import (
-	"bufio"
+	"bytes"
 	"log"
 	"os"
 )
 
 func ReadLines(name string) []string {
-	f, err := os.Open(name)
+	data, err := os.ReadFile(name)
 	if err != nil {
-		log.Fatal("ReadLines: ", err)
+		log.Fatal("ReadFile: ", err)
 	}
-	defer f.Close()
-	r := bufio.NewReader(f)
 
-	l := []string{}
-	for {
-		line, _, err := r.ReadLine()
-		if line == nil || err != nil {
-			break
+	var lines []string
+	for _, l := range bytes.Split(data, []byte("\n")) {
+		if len(l) > 0 {
+			lines = append(lines, string(l))
 		}
-
-		fn := string(line)
-		l = append(l, fn)
-	}
-	if len(l) == 0 {
-		log.Fatal("no files added")
 	}
-
-	return l
+	return lines
 }

benchmark/read_test.go

@@ -5,54 +5,130 @@
 package benchmark
 
 import (
+	"bytes"
+	"flag"
 	"fmt"
 	"os"
 	"os/exec"
 	"testing"
+	"time"
 
+	"github.com/hanwen/go-fuse/v2/fs"
 	"github.com/hanwen/go-fuse/v2/internal/testutil"
-	"golang.org/x/sync/errgroup"
 )
 
-func BenchmarkGoFuseRead(b *testing.B) {
-	fs := &readFS{}
-	wd, clean := setupFs(fs, b.N)
-	defer clean()
-
-	jobs := 32
-	blockSize := 64 * 1024
+func BenchmarkGoFuseMemoryRead(b *testing.B) {
+	root := &readFS{}
+	mnt := setupFS(root, b.N, b)
+	benchmarkRead(mnt, b, 32, "direct")
+}
 
-	cmds := make([]*exec.Cmd, jobs)
+const blockSize = 64 * 1024
 
-	for i := 0; i < jobs; i++ {
-		cmds[i] = exec.Command("dd",
-			fmt.Sprintf("if=%s/foo.txt", wd),
-			"iflag=direct",
+func benchmarkRead(mnt string, b *testing.B, readers int, ddflag string) {
+	var cmds []*exec.Cmd
+	for i := 0; i < readers; i++ {
+		cmd := exec.Command("dd",
+			fmt.Sprintf("if=%s/foo.txt", mnt),
 			"of=/dev/null",
 			fmt.Sprintf("bs=%d", blockSize),
 			fmt.Sprintf("count=%d", b.N))
+		if ddflag != "" {
+			cmd.Args = append(cmd.Args, "iflag="+ddflag)
+		}
 		if testutil.VerboseTest() {
-			cmds[i].Stdout = os.Stdout
-			cmds[i].Stderr = os.Stderr
+			cmd.Stderr = os.Stderr
+			cmd.Stdout = os.Stdout
+		} else {
+			buf := &bytes.Buffer{}
+			cmd.Stderr = buf
+			cmd.Stdout = buf
 		}
+		cmds = append(cmds, cmd)
 	}
 
-	b.SetBytes(int64(jobs * blockSize))
+	b.SetBytes(int64(readers * blockSize))
 	b.ReportAllocs()
 	b.ResetTimer()
 
-	var eg errgroup.Group
+	result := make(chan error, readers)
+	for _, cmd := range cmds {
+		go func(cmd *exec.Cmd) {
+			err := cmd.Run()
+			if buf, ok := cmd.Stdout.(*bytes.Buffer); ok && err != nil {
+				err = fmt.Errorf("%v: output=%s", err, buf.String())
+			}
+			result <- err
+		}(cmd)
+	}
+	failures := 0
+	for range cmds {
+		if err := <-result; err != nil {
+			b.Errorf("dd failed: %v", err)
+			failures++
+		}
+	}
+	if failures > 0 {
+		b.Errorf("%d out of %d commands", failures, readers)
+	}
+	b.StopTimer()
+}
+
+func BenchmarkGoFuseFDRead(b *testing.B) {
+	orig := b.TempDir()
+	fn := orig + "/foo.txt"
 
-	for i := 0; i < jobs; i++ {
-		i := i
-		eg.Go(func() error {
-			return cmds[i].Run()
-		})
+	data := bytes.Repeat([]byte{42}, blockSize*b.N)
+	if err := os.WriteFile(fn, data, 0666); err != nil {
+		b.Fatal(err)
+	}
+	root, err := fs.NewLoopbackRoot(orig)
+	if err != nil {
+		b.Fatal(err)
 	}
+	mnt := setupFS(root, b.N, b)
+	benchmarkRead(mnt, b, 1, "")
+}
 
-	if err := eg.Wait(); err != nil {
-		b.Fatalf("dd failed: %v", err)
+var libfusePath = flag.String("passthrough_hp", "", "path to libfuse's passthrough_hp")
+
+func BenchmarkLibfuseHP(b *testing.B) {
+	orig := b.TempDir()
+	mnt := b.TempDir()
+	if *libfusePath == "" {
+		b.Skip("must set --passthrough_hp")
 	}
 
-	b.StopTimer()
+	origFN := orig + "/foo.txt"
+	data := bytes.Repeat([]byte{42}, blockSize*b.N)
+	if err := os.WriteFile(origFN, data, 0666); err != nil {
+		b.Fatal(err)
+	}
+	fn := mnt + "/foo.txt"
+	cmd := exec.Command(*libfusePath, "--foreground")
+	if testutil.VerboseTest() {
+		cmd.Args = append(cmd.Args, "--debug", "--debug-fuse")
+	}
+	cmd.Args = append(cmd.Args, orig, mnt)
+	cmd.Stderr = os.Stderr
+	cmd.Stdout = os.Stdout
+
+	if err := cmd.Start(); err != nil {
+		b.Fatal(err)
+	}
+	b.Cleanup(func() { exec.Command("fusermount", "-u", mnt).Run() })
+
+	dt := time.Millisecond
+	for {
+		if _, err := os.Stat(fn); err == nil {
+			break
+		}
+		time.Sleep(dt)
+		dt *= 2
+		if dt > time.Second {
+			b.Fatal("file did not appear")
+		}
+	}
+
+	benchmarkRead(mnt, b, 1, "")
 }

benchmark/stat_test.go

@@ -22,22 +22,19 @@ import (
 	"github.com/hanwen/go-fuse/v2/internal/testutil"
 )
 
-func setupFs(node fs.InodeEmbedder, N int) (string, func()) {
+func setupFS(node fs.InodeEmbedder, N int, tb testing.TB) string {
 	opts := &fs.Options{}
 	opts.Debug = testutil.VerboseTest()
-	mountPoint, err := os.MkdirTemp("", "")
-	if err != nil {
-		log.Panicf("TempDir: %v", err)
-	}
+	mountPoint := tb.TempDir()
 	server, err := fs.Mount(mountPoint, node, opts)
 	if err != nil {
-		log.Panicf("cannot mount %v", err)
+		tb.Fatalf("cannot mount %v", err)
 	}
 	lmap := NewLatencyMap()
 	if testutil.VerboseTest() {
 		server.RecordLatencies(lmap)
 	}
-	return mountPoint, func() {
+	tb.Cleanup(func() {
 		if testutil.VerboseTest() {
 			var total time.Duration
 			for _, n := range []string{"LOOKUP", "GETATTR", "OPENDIR", "READDIR",
@@ -57,7 +54,8 @@ func setupFs(node fs.InodeEmbedder, N int) (string, func()) {
 		if err != nil {
 			log.Println("error during unmount", err)
 		}
-	}
+	})
+	return mountPoint
 }
 
 func TestNewStatFs(t *testing.T) {
@@ -68,8 +66,7 @@ func TestNewStatFs(t *testing.T) {
 		fs.AddFile(n, fuse.Attr{Mode: syscall.S_IFREG})
 	}
 
-	wd, clean := setupFs(fs, 1)
-	defer clean()
+	wd := setupFS(fs, 1, t)
 
 	names, err := ioutil.ReadDir(wd)
 	if err != nil {
@@ -121,15 +118,14 @@ func BenchmarkGoFuseStat(b *testing.B) {
 		fs.AddFile(fn, fuse.Attr{Mode: syscall.S_IFREG})
 	}
 
-	wd, clean := setupFs(fs, b.N)
-	defer clean()
+	mnt := setupFS(fs, b.N, b)
 
 	for i, l := range files {
-		files[i] = filepath.Join(wd, l)
+		files[i] = filepath.Join(mnt, l)
 	}
 
 	threads := runtime.GOMAXPROCS(0)
-	if err := TestingBOnePass(b, threads, fileList, wd); err != nil {
+	if err := TestingBOnePass(b, threads, fileList, mnt); err != nil {
 		b.Fatalf("TestingBOnePass %v8", err)
 	}
 }
@@ -157,12 +153,11 @@ func BenchmarkGoFuseReaddir(b *testing.B) {
 		dirSet[filepath.Dir(fn)] = struct{}{}
 	}
 
-	wd, clean := setupFs(fs, b.N)
-	defer clean()
+	mnt := setupFS(fs, b.N, b)
 
 	var dirs []string
 	for dir := range dirSet {
-		dirs = append(dirs, filepath.Join(wd, dir))
+		dirs = append(dirs, filepath.Join(mnt, dir))
 	}
 	b.StartTimer()
 	todo := b.N

fs/api.go

@@ -9,8 +9,8 @@
 // To create a file system, you should first define types for the
 // nodes of the file system tree.
 //
-//	struct myNode {
-//	   fs.Inode
+//	type myNode struct {
+//		fs.Inode
 //	}
 //
 //	// Node types must be InodeEmbedders
@@ -20,10 +20,10 @@
 //	var _ = (fs.NodeLookuper)((*myNode)(nil))
 //
 //	func (n *myNode) Lookup(ctx context.Context, name string, out *fuse.EntryOut) (*Inode, syscall.Errno) {
-//	  ops := myNode{}
-//        out.Mode = 0755
-//        out.Size = 42
-//	  return n.NewInode(ctx, &ops, fs.StableAttr{Mode: syscall.S_IFREG}), 0
+//		ops := myNode{}
+//		out.Mode = 0755
+//		out.Size = 42
+//		return n.NewInode(ctx, &ops, fs.StableAttr{Mode: syscall.S_IFREG}), 0
 //	}
 //
 // The method names are inspired on the system call names, so we have
@@ -568,6 +568,17 @@ type NodeRenamer interface {
 type FileHandle interface {
 }
 
+// FilePassthroughFder is a file backed by a physical
+// file. PassthroughFd should return an open file descriptor (and
+// true), and the kernel will execute read/write operations directly
+// on the backing file, bypassing the FUSE process. This function will
+// be called once when processing the Create or Open operation, so
+// there is no concern about concurrent access to the Fd. If the
+// function returns false, passthrough will not be used for this file.
+type FilePassthroughFder interface {
+	PassthroughFd() (int, bool)
+}
+
 // See NodeReleaser.
 type FileReleaser interface {
 	Release(ctx context.Context) syscall.Errno

fs/constants_darwin.go

-// Copyright 2019 the Go-FUSE Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package fs
-
-import "syscall"
-
-// ENOATTR indicates that an extended attribute was not present.
-var ENOATTR = syscall.ENOATTR

fs/constants_linux.go

@@ -4,7 +4,7 @@
 
 package fs
 
-import "syscall"
+import "golang.org/x/sys/unix"
 
 // ENOATTR indicates that an extended attribute was not present.
-var ENOATTR = syscall.ENODATA
+const ENOATTR = unix.ENODATA

fs/constants_unix.go

+//go:build !linux
+
+package fs
+
+import "golang.org/x/sys/unix"
+
+const ENOATTR = unix.ENOATTR

fs/dir_test.go

@@ -45,7 +45,7 @@ func (ds *errDirStream) Next() (fuse.DirEntry, syscall.Errno) {
 			Mode: fuse.S_IFREG,
 			Name: "last",
 			Ino:  3,
-		}, syscall.EKEYEXPIRED
+		}, syscall.EBADMSG
 	}
 
 	panic("boom")
@@ -76,9 +76,14 @@ func TestDirStreamError(t *testing.T) {
 				} else if e.Name != "first" {
 					t.Errorf("got %q want 'first'", e.Name)
 				}
-
-				if _, errno := ds.Next(); errno != syscall.EKEYEXPIRED {
-					t.Errorf("got errno %v, want EKEYEXPIRED", errno)
+				// Here we need choose a errno to test if errno could be passed and handled
+				// correctly by the fuse library. To build the test on different platform,
+				// an errno which defined on each platform should be chosen. And if the
+				// chosen integer number is not a valid errno, the fuse in kernel would refuse
+				// and throw an error, which is observed on Linux.
+				// Here we choose to use EBADMSG, which is defined on multiple Unix-like OSes.
+				if _, errno := ds.Next(); errno != syscall.EBADMSG {
+					t.Errorf("got errno %v, want EBADMSG", errno)
 				}
 			})
 	}

fs/dirstream_darwin.go

@@ -12,6 +12,16 @@ import (
 	"github.com/hanwen/go-fuse/v2/fuse"
 )
 
+// Like syscall.Dirent, but without the [256]byte name.
+type dirent struct {
+	Ino    uint64
+	Off    int64
+	Reclen uint16
+	Namlen uint16
+	Type   uint8
+	Name   [1]uint8 // align to 4 bytes for 32 bits.
+}
+
 func NewLoopbackDirStream(nm string) (DirStream, syscall.Errno) {
 	f, err := os.Open(nm)
 	if err != nil {

fs/dirstream_freebsd.go

+package fs
+
+// Like syscall.Dirent, but without the [256]byte name.
+type dirent struct {
+	Ino    uint64
+	Off    int64
+	Reclen uint16
+	Type   uint8
+	Pad0   uint8
+	Namlen uint16
+	Pad1   uint16
+	Name   [1]uint8 // align to 4 bytes for 32 bits.
+}

fs/dirstream_linux.go

-// Copyright 2019 the Go-FUSE Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
 package fs
 
-import (
-	"sync"
-	"syscall"
-	"unsafe"
-
-	"github.com/hanwen/go-fuse/v2/fuse"
-)
-
-type loopbackDirStream struct {
-	buf       []byte
-	todo      []byte
-	todoErrno syscall.Errno
-
-	// Protects fd so we can guard against double close
-	mu sync.Mutex
-	fd int
-}
-
-// NewLoopbackDirStream open a directory for reading as a DirStream
-func NewLoopbackDirStream(name string) (DirStream, syscall.Errno) {
-	fd, err := syscall.Open(name, syscall.O_DIRECTORY, 0755)
-	if err != nil {
-		return nil, ToErrno(err)
-	}
-
-	ds := &loopbackDirStream{
-		buf: make([]byte, 4096),
-		fd:  fd,
-	}
-
-	if err := ds.load(); err != 0 {
-		ds.Close()
-		return nil, err
-	}
-	return ds, OK
-}
-
-func (ds *loopbackDirStream) Close() {
-	ds.mu.Lock()
-	defer ds.mu.Unlock()
-	if ds.fd != -1 {
-		syscall.Close(ds.fd)
-		ds.fd = -1
-	}
-}
-
-func (ds *loopbackDirStream) HasNext() bool {
-	ds.mu.Lock()
-	defer ds.mu.Unlock()
-	return len(ds.todo) > 0 || ds.todoErrno != 0
-}
-
 // Like syscall.Dirent, but without the [256]byte name.
 type dirent struct {
 	Ino    uint64
@@ -64,50 +8,3 @@ type dirent struct {
 	Type   uint8
 	Name   [1]uint8 // align to 4 bytes for 32 bits.
 }
-
-func (ds *loopbackDirStream) Next() (fuse.DirEntry, syscall.Errno) {
-	ds.mu.Lock()
-	defer ds.mu.Unlock()
-
-	if ds.todoErrno != 0 {
-		return fuse.DirEntry{}, ds.todoErrno
-	}
-
-	// We can't use syscall.Dirent here, because it declares a
-	// [256]byte name, which may run beyond the end of ds.todo.
-	// when that happens in the race detector, it causes a panic
-	// "converted pointer straddles multiple allocations"
-	de := (*dirent)(unsafe.Pointer(&ds.todo[0]))
-
-	nameBytes := ds.todo[unsafe.Offsetof(dirent{}.Name):de.Reclen]
-	ds.todo = ds.todo[de.Reclen:]
-
-	// After the loop, l contains the index of the first '\0'.
-	l := 0
-	for l = range nameBytes {
-		if nameBytes[l] == 0 {
-			break
-		}
-	}
-	nameBytes = nameBytes[:l]
-	result := fuse.DirEntry{
-		Ino:  de.Ino,
-		Mode: (uint32(de.Type) << 12),
-		Name: string(nameBytes),
-	}
-	return result, ds.load()
-}
-
-func (ds *loopbackDirStream) load() syscall.Errno {
-	if len(ds.todo) > 0 {
-		return OK
-	}
-
-	n, err := syscall.Getdents(ds.fd, ds.buf)
-	if n < 0 {
-		n = 0
-	}
-	ds.todo = ds.buf[:n]
-	ds.todoErrno = ToErrno(err)
-	return OK
-}

fs/dirstream_unix.go

+//go:build !darwin
+
+// Copyright 2019 the Go-FUSE Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package fs
+
+import (
+	"sync"
+	"syscall"
+	"unsafe"
+
+	"github.com/hanwen/go-fuse/v2/fuse"
+	"golang.org/x/sys/unix"
+)
+
+type loopbackDirStream struct {
+	buf       []byte
+	todo      []byte
+	todoErrno syscall.Errno
+
+	// Protects fd so we can guard against double close
+	mu sync.Mutex
+	fd int
+}
+
+// NewLoopbackDirStream open a directory for reading as a DirStream
+func NewLoopbackDirStream(name string) (DirStream, syscall.Errno) {
+	fd, err := syscall.Open(name, syscall.O_DIRECTORY, 0755)
+	if err != nil {
+		return nil, ToErrno(err)
+	}
+
+	ds := &loopbackDirStream{
+		buf: make([]byte, 4096),
+		fd:  fd,
+	}
+
+	if err := ds.load(); err != 0 {
+		ds.Close()
+		return nil, err
+	}
+	return ds, OK
+}
+
+func (ds *loopbackDirStream) Close() {
+	ds.mu.Lock()
+	defer ds.mu.Unlock()
+	if ds.fd != -1 {
+		syscall.Close(ds.fd)
+		ds.fd = -1
+	}
+}
+
+func (ds *loopbackDirStream) HasNext() bool {
+	ds.mu.Lock()
+	defer ds.mu.Unlock()
+	return len(ds.todo) > 0 || ds.todoErrno != 0
+}
+
+func (ds *loopbackDirStream) Next() (fuse.DirEntry, syscall.Errno) {
+	ds.mu.Lock()
+	defer ds.mu.Unlock()
+
+	if ds.todoErrno != 0 {
+		return fuse.DirEntry{}, ds.todoErrno
+	}
+
+	// We can't use syscall.Dirent here, because it declares a
+	// [256]byte name, which may run beyond the end of ds.todo.
+	// when that happens in the race detector, it causes a panic
+	// "converted pointer straddles multiple allocations"
+	de := (*dirent)(unsafe.Pointer(&ds.todo[0]))
+
+	nameBytes := ds.todo[unsafe.Offsetof(dirent{}.Name):de.Reclen]
+	ds.todo = ds.todo[de.Reclen:]
+
+	// After the loop, l contains the index of the first '\0'.
+	l := 0
+	for l = range nameBytes {
+		if nameBytes[l] == 0 {
+			break
+		}
+	}
+	nameBytes = nameBytes[:l]
+	result := fuse.DirEntry{
+		Ino:  de.Ino,
+		Mode: (uint32(de.Type) << 12),
+		Name: string(nameBytes),
+	}
+	return result, ds.load()
+}
+
+func (ds *loopbackDirStream) load() syscall.Errno {
+	if len(ds.todo) > 0 {
+		return OK
+	}
+
+	n, err := unix.Getdents(ds.fd, ds.buf)
+	if n < 0 {
+		n = 0
+	}
+	ds.todo = ds.buf[:n]
+	ds.todoErrno = ToErrno(err)
+	return OK
+}

fs/files.go

@@ -7,12 +7,10 @@ package fs
 import (
 	"context"
 	"sync"
-
-	//	"time"
-
 	"syscall"
 
 	"github.com/hanwen/go-fuse/v2/fuse"
+	"github.com/hanwen/go-fuse/v2/internal/fallocate"
 	"golang.org/x/sys/unix"
 )
 
@@ -42,6 +40,14 @@ var _ = (FileFlusher)((*loopbackFile)(nil))
 var _ = (FileFsyncer)((*loopbackFile)(nil))
 var _ = (FileSetattrer)((*loopbackFile)(nil))
 var _ = (FileAllocater)((*loopbackFile)(nil))
+var _ = (FilePassthroughFder)((*loopbackFile)(nil))
+
+func (f *loopbackFile) PassthroughFd() (int, bool) {
+	// This Fd is not accessed concurrently, but lock anyway for uniformity.
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.fd, true
+}
 
 func (f *loopbackFile) Read(ctx context.Context, buf []byte, off int64) (res fuse.ReadResult, errno syscall.Errno) {
 	f.mu.Lock()
@@ -241,3 +247,13 @@ func (f *loopbackFile) Lseek(ctx context.Context, off uint64, whence uint32) (ui
 	n, err := unix.Seek(f.fd, int64(off), int(whence))
 	return uint64(n), ToErrno(err)
 }
+
+func (f *loopbackFile) Allocate(ctx context.Context, off uint64, sz uint64, mode uint32) syscall.Errno {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	err := fallocate.Fallocate(f.fd, mode, int64(off), int64(sz))
+	if err != nil {
+		return ToErrno(err)
+	}
+	return OK
+}

fs/files_darwin.go

@@ -4,7 +4,29 @@
 
 package fs
 
-import "github.com/hanwen/go-fuse/v2/fuse"
+import (
+	"context"
+	"syscall"
+	"time"
+
+	"github.com/hanwen/go-fuse/v2/fuse"
+	"github.com/hanwen/go-fuse/v2/internal/utimens"
+)
 
 func setBlocks(out *fuse.Attr) {
 }
+
+// MacOS before High Sierra lacks utimensat() and UTIME_OMIT.
+// We emulate using utimes() and extra Getattr() calls.
+func (f *loopbackFile) utimens(a *time.Time, m *time.Time) syscall.Errno {
+	var attr fuse.AttrOut
+	if a == nil || m == nil {
+		errno := f.Getattr(context.Background(), &attr)
+		if errno != 0 {
+			return errno
+		}
+	}
+	tv := utimens.Fill(a, m, &attr.Attr)
+	err := syscall.Futimes(int(f.fd), tv)
+	return ToErrno(err)
+}

fs/files_freebsd.go

+package fs
+
+import "github.com/hanwen/go-fuse/v2/fuse"
+
+func setBlocks(out *fuse.Attr) {
+}

fs/files_linux.go

@@ -5,32 +5,9 @@
 package fs
 
 import (
-	"context"
-	"syscall"
-	"time"
-
 	"github.com/hanwen/go-fuse/v2/fuse"
 )
 
-func (f *loopbackFile) Allocate(ctx context.Context, off uint64, sz uint64, mode uint32) syscall.Errno {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-	err := syscall.Fallocate(f.fd, mode, int64(off), int64(sz))
-	if err != nil {
-		return ToErrno(err)
-	}
-	return OK
-}
-
-// Utimens - file handle based version of loopbackFileSystem.Utimens()
-func (f *loopbackFile) utimens(a *time.Time, m *time.Time) syscall.Errno {
-	var ts [2]syscall.Timespec
-	ts[0] = fuse.UtimeToTimespec(a)
-	ts[1] = fuse.UtimeToTimespec(m)
-	err := futimens(int(f.fd), &ts)
-	return ToErrno(err)
-}
-
 func setBlocks(out *fuse.Attr) {
 	if out.Blksize > 0 {
 		return

fs/syscall_linux.go → fs/files_unix.go

-// Copyright 2016 the Go-FUSE Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
+//go:build !darwin
 
 package fs
 
 import (
 	"syscall"
+	"time"
 	"unsafe"
+
+	"github.com/hanwen/go-fuse/v2/fuse"
 )
 
+// Utimens - file handle based version of loopbackFileSystem.Utimens()
+func (f *loopbackFile) utimens(a *time.Time, m *time.Time) syscall.Errno {
+	var ts [2]syscall.Timespec
+	ts[0] = fuse.UtimeToTimespec(a)
+	ts[1] = fuse.UtimeToTimespec(m)
+	err := futimens(int(f.fd), &ts)
+	return ToErrno(err)
+}
+
 // futimens - futimens(3) calls utimensat(2) with "pathname" set to null and
 // "flags" set to zero
 func futimens(fd int, times *[2]syscall.Timespec) (err error) {

fs/inode.go

@@ -69,6 +69,11 @@ type Inode struct {
 	// protected by bridge.mu
 	openFiles []uint32
 
+	// backing files, protected by bridge.mu
+	backingIDRefcount int
+	backingID         int32
+	backingFd         int
+
 	// mu protects the following mutable fields. When locking
 	// multiple Inodes, locks must be acquired using
 	// lockNodes/unlockNodes

fs/loopback.go

@@ -12,6 +12,7 @@ import (
 
 	"github.com/hanwen/go-fuse/v2/fuse"
 	"github.com/hanwen/go-fuse/v2/internal/renameat"
+	"golang.org/x/sys/unix"
 )
 
 // LoopbackRoot holds the parameters for creating a new loopback
@@ -123,7 +124,7 @@ var _ = (NodeMknoder)((*LoopbackNode)(nil))
 
 func (n *LoopbackNode) Mknod(ctx context.Context, name string, mode, rdev uint32, out *fuse.EntryOut) (*Inode, syscall.Errno) {
 	p := filepath.Join(n.path(), name)
-	err := syscall.Mknod(p, mode, int(rdev))
+	err := syscall.Mknod(p, mode, intDev(rdev))
 	if err != nil {
 		return nil, ToErrno(err)
 	}
@@ -402,20 +403,23 @@ func (n *LoopbackNode) Setattr(ctx context.Context, f FileHandle, in *fuse.SetAt
 		atime, aok := in.GetATime()
 
 		if mok || aok {
-
-			ap := &atime
-			mp := &mtime
-			if !aok {
-				ap = nil
+			ta := unix.Timespec{Nsec: unix_UTIME_OMIT}
+			tm := unix.Timespec{Nsec: unix_UTIME_OMIT}
+			var err error
+			if aok {
+				ta, err = unix.TimeToTimespec(atime)
+				if err != nil {
+					return ToErrno(err)
+				}
 			}
-			if !mok {
-				mp = nil
+			if mok {
+				tm, err = unix.TimeToTimespec(mtime)
+				if err != nil {
+					return ToErrno(err)
+				}
 			}
-			var ts [2]syscall.Timespec
-			ts[0] = fuse.UtimeToTimespec(ap)
-			ts[1] = fuse.UtimeToTimespec(mp)
-
-			if err := syscall.UtimesNano(p, ts[:]); err != nil {
+			ts := []unix.Timespec{ta, tm}
+			if err := unix.UtimesNanoAt(unix.AT_FDCWD, p, ts, unix.AT_SYMLINK_NOFOLLOW); err != nil {
 				return ToErrno(err)
 			}
 		}
@@ -441,6 +445,46 @@ func (n *LoopbackNode) Setattr(ctx context.Context, f FileHandle, in *fuse.SetAt
 	return OK
 }
 
+var _ = (NodeGetxattrer)((*LoopbackNode)(nil))
+
+func (n *LoopbackNode) Getxattr(ctx context.Context, attr string, dest []byte) (uint32, syscall.Errno) {
+	sz, err := unix.Lgetxattr(n.path(), attr, dest)
+	return uint32(sz), ToErrno(err)
+}
+
+var _ = (NodeSetxattrer)((*LoopbackNode)(nil))
+
+func (n *LoopbackNode) Setxattr(ctx context.Context, attr string, data []byte, flags uint32) syscall.Errno {
+	err := unix.Lsetxattr(n.path(), attr, data, int(flags))
+	return ToErrno(err)
+}
+
+var _ = (NodeRemovexattrer)((*LoopbackNode)(nil))
+
+func (n *LoopbackNode) Removexattr(ctx context.Context, attr string) syscall.Errno {
+	err := unix.Lremovexattr(n.path(), attr)
+	return ToErrno(err)
+}
+
+var _ = (NodeCopyFileRanger)((*LoopbackNode)(nil))
+
+func (n *LoopbackNode) CopyFileRange(ctx context.Context, fhIn FileHandle,
+	offIn uint64, out *Inode, fhOut FileHandle, offOut uint64,
+	len uint64, flags uint64) (uint32, syscall.Errno) {
+	lfIn, ok := fhIn.(*loopbackFile)
+	if !ok {
+		return 0, unix.ENOTSUP
+	}
+	lfOut, ok := fhOut.(*loopbackFile)
+	if !ok {
+		return 0, unix.ENOTSUP
+	}
+	signedOffIn := int64(offIn)
+	signedOffOut := int64(offOut)
+	doCopyFileRange(lfIn.fd, signedOffIn, lfOut.fd, signedOffOut, int(len), int(flags))
+	return 0, syscall.ENOSYS
+}
+
 // NewLoopbackRoot returns a root node for a loopback file system whose
 // root is at the given root. This node implements all NodeXxxxer
 // operations available.

fs/loopback_darwin.go

@@ -8,86 +8,11 @@
 package fs
 
 import (
-	"context"
 	"syscall"
 	"time"
-	"unsafe"
-
-	"github.com/hanwen/go-fuse/v2/fuse"
-	"github.com/hanwen/go-fuse/v2/internal/utimens"
 )
 
-var _ = (NodeGetxattrer)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) Getxattr(ctx context.Context, attr string, dest []byte) (uint32, syscall.Errno) {
-	return 0, syscall.ENOSYS
-}
-
-var _ = (NodeSetxattrer)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) Setxattr(ctx context.Context, attr string, data []byte, flags uint32) syscall.Errno {
-	return syscall.ENOSYS
-}
-
-var _ = (NodeRemovexattrer)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) Removexattr(ctx context.Context, attr string) syscall.Errno {
-	return syscall.ENOSYS
-}
-
-var _ = (NodeListxattrer)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) Listxattr(ctx context.Context, dest []byte) (uint32, syscall.Errno) {
-	return 0, syscall.ENOSYS
-}
-
-func (f *loopbackFile) Allocate(ctx context.Context, off uint64, sz uint64, mode uint32) syscall.Errno {
-	// TODO: Handle `mode` parameter.
-
-	// From `man fcntl` on OSX:
-	//     The F_PREALLOCATE command operates on the following structure:
-	//
-	//             typedef struct fstore {
-	//                 u_int32_t fst_flags;      /* IN: flags word */
-	//                 int       fst_posmode;    /* IN: indicates offset field */
-	//                 off_t     fst_offset;     /* IN: start of the region */
-	//                 off_t     fst_length;     /* IN: size of the region */
-	//                 off_t     fst_bytesalloc; /* OUT: number of bytes allocated */
-	//             } fstore_t;
-	//
-	//     The flags (fst_flags) for the F_PREALLOCATE command are as follows:
-	//
-	//           F_ALLOCATECONTIG   Allocate contiguous space.
-	//
-	//           F_ALLOCATEALL      Allocate all requested space or no space at all.
-	//
-	//     The position modes (fst_posmode) for the F_PREALLOCATE command indicate how to use the offset field.  The modes are as fol-
-	//     lows:
-	//
-	//           F_PEOFPOSMODE   Allocate from the physical end of file.
-	//
-	//           F_VOLPOSMODE    Allocate from the volume offset.
-
-	k := struct {
-		Flags      uint32 // u_int32_t
-		Posmode    int64  // int
-		Offset     int64  // off_t
-		Length     int64  // off_t
-		Bytesalloc int64  // off_t
-	}{
-		0,
-		0,
-		int64(off),
-		int64(sz),
-		0,
-	}
-
-	// Linux version for reference:
-	// err := syscall.Fallocate(int(f.File.Fd()), mode, int64(off), int64(sz))
-	_, _, errno := syscall.Syscall(syscall.SYS_FCNTL, uintptr(f.fd), uintptr(syscall.F_PREALLOCATE), uintptr(unsafe.Pointer(&k)))
-
-	return errno
-}
+const unix_UTIME_OMIT = 0x0
 
 // timeToTimeval - Convert time.Time to syscall.Timeval
 //
@@ -101,25 +26,11 @@ func timeToTimeval(t *time.Time) syscall.Timeval {
 	return tv
 }
 
-// MacOS before High Sierra lacks utimensat() and UTIME_OMIT.
-// We emulate using utimes() and extra Getattr() calls.
-func (f *loopbackFile) utimens(a *time.Time, m *time.Time) syscall.Errno {
-	var attr fuse.AttrOut
-	if a == nil || m == nil {
-		errno := f.Getattr(context.Background(), &attr)
-		if errno != 0 {
-			return errno
-		}
-	}
-	tv := utimens.Fill(a, m, &attr.Attr)
-	err := syscall.Futimes(int(f.fd), tv)
-	return ToErrno(err)
+func doCopyFileRange(fdIn int, offIn int64, fdOut int, offOut int64,
+	len int, flags int) (uint32, syscall.Errno) {
+	return 0, syscall.ENOSYS
 }
 
-var _ = (NodeCopyFileRanger)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) CopyFileRange(ctx context.Context, fhIn FileHandle,
-	offIn uint64, out *Inode, fhOut FileHandle, offOut uint64,
-	len uint64, flags uint64) (uint32, syscall.Errno) {
-	return 0, syscall.ENOSYS
+func intDev(dev uint32) int {
+	return int(dev)
 }

fs/loopback_freebsd.go

+package fs
+
+import (
+	"context"
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+const unix_UTIME_OMIT = unix.UTIME_OMIT
+
+// FreeBSD has added copy_file_range(2) since FreeBSD 12. However,
+// golang.org/x/sys/unix hasn't add corresponding syscall constant or
+// wrap function. Here we define the syscall constant until sys/unix
+// provides.
+const sys_COPY_FILE_RANGE = 569
+
+// TODO: replace the manual syscall when sys/unix provides CopyFileRange
+// for FreeBSD
+func doCopyFileRange(fdIn int, offIn int64, fdOut int, offOut int64,
+	len int, flags int) (uint32, syscall.Errno) {
+	count, _, errno := unix.Syscall6(sys_COPY_FILE_RANGE,
+		uintptr(fdIn), uintptr(offIn), uintptr(fdOut), uintptr(offOut),
+		uintptr(len), uintptr(flags),
+	)
+	return uint32(count), errno
+}
+
+func intDev(dev uint32) uint64 {
+	return uint64(dev)
+}
+
+// BSDs syscall use different convention of data buf retrieved
+// through syscall `unix.Listxattr`.
+// Ref: extattr_list_file(2)
+func retrieveAttrName(buf []byte) [][]byte {
+	var attrList [][]byte
+	for p := 0; p < len(buf); {
+		attrNameLen := int(buf[p])
+		p++
+		attrName := buf[p : p+attrNameLen]
+		attrList = append(attrList, attrName)
+		p += attrNameLen
+	}
+	return attrList
+}
+
+// Since FUSE on FreeBSD expect Linux flavor data format of
+// listxattr, we should reconstruct it with data returned by
+// FreeBSD's syscall. And here we have added a "user." prefix
+// to put them under "user" namespace, which is readable and
+// writable for normal user, for a userspace implemented FS.
+func rebuildAttrBuf(attrList [][]byte) []byte {
+	ret := make([]byte, 0)
+	for _, attrName := range attrList {
+		nsAttrName := append([]byte("user."), attrName...)
+		ret = append(ret, nsAttrName...)
+		ret = append(ret, 0x0)
+	}
+	return ret
+}
+
+var _ = (NodeListxattrer)((*LoopbackNode)(nil))
+
+func (n *LoopbackNode) Listxattr(ctx context.Context, dest []byte) (uint32, syscall.Errno) {
+	// In order to simulate same data format as Linux does,
+	// and the size of returned buf is required to match, we must
+	// call unix.Llistxattr twice.
+	sz, err := unix.Llistxattr(n.path(), nil)
+	if err != nil {
+		return uint32(sz), ToErrno(err)
+	}
+	rawBuf := make([]byte, sz)
+	sz, err = unix.Llistxattr(n.path(), rawBuf)
+	if err != nil {
+		return uint32(sz), ToErrno(err)
+	}
+	attrList := retrieveAttrName(rawBuf)
+	rebuiltBuf := rebuildAttrBuf(attrList)
+	sz = len(rebuiltBuf)
+	if len(dest) != 0 {
+		// When len(dest) is 0, which means that caller wants to get
+		// the size. If len(dest) is less than len(rebuiltBuf), but greater
+		// than 0 dest will be also filled with data from rebuiltBuf,
+		// but truncated to len(dest). copy() function will do the same.
+		// And this behaviour is same as FreeBSD's syscall extattr_list_file(2).
+		sz = copy(dest, rebuiltBuf)
+	}
+	return uint32(sz), ToErrno(err)
+}

fs/loopback_linux.go

@@ -8,56 +8,19 @@
 package fs
 
 import (
-	"context"
 	"syscall"
 
 	"golang.org/x/sys/unix"
 )
 
-var _ = (NodeGetxattrer)((*LoopbackNode)(nil))
+const unix_UTIME_OMIT = unix.UTIME_OMIT
 
-func (n *LoopbackNode) Getxattr(ctx context.Context, attr string, dest []byte) (uint32, syscall.Errno) {
-	sz, err := unix.Lgetxattr(n.path(), attr, dest)
-	return uint32(sz), ToErrno(err)
-}
-
-var _ = (NodeSetxattrer)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) Setxattr(ctx context.Context, attr string, data []byte, flags uint32) syscall.Errno {
-	err := unix.Lsetxattr(n.path(), attr, data, int(flags))
-	return ToErrno(err)
-}
-
-var _ = (NodeRemovexattrer)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) Removexattr(ctx context.Context, attr string) syscall.Errno {
-	err := unix.Lremovexattr(n.path(), attr)
-	return ToErrno(err)
-}
-
-var _ = (NodeListxattrer)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) Listxattr(ctx context.Context, dest []byte) (uint32, syscall.Errno) {
-	sz, err := unix.Llistxattr(n.path(), dest)
-	return uint32(sz), ToErrno(err)
+func doCopyFileRange(fdIn int, offIn int64, fdOut int, offOut int64,
+	len int, flags int) (uint32, syscall.Errno) {
+	count, err := unix.CopyFileRange(fdIn, &offIn, fdOut, &offOut, len, flags)
+	return uint32(count), ToErrno(err)
 }
 
-var _ = (NodeCopyFileRanger)((*LoopbackNode)(nil))
-
-func (n *LoopbackNode) CopyFileRange(ctx context.Context, fhIn FileHandle,
-	offIn uint64, out *Inode, fhOut FileHandle, offOut uint64,
-	len uint64, flags uint64) (uint32, syscall.Errno) {
-	lfIn, ok := fhIn.(*loopbackFile)
-	if !ok {
-		return 0, syscall.ENOTSUP
-	}
-	lfOut, ok := fhOut.(*loopbackFile)
-	if !ok {
-		return 0, syscall.ENOTSUP
-	}
-
-	signedOffIn := int64(offIn)
-	signedOffOut := int64(offOut)
-	count, err := unix.CopyFileRange(lfIn.fd, &signedOffIn, lfOut.fd, &signedOffOut, int(len), int(flags))
-	return uint32(count), ToErrno(err)
+func intDev(dev uint32) int {
+	return int(dev)
 }

fs/loopback_linux_test.go

@@ -45,63 +45,6 @@ func TestRenameNoOverwrite(t *testing.T) {
 	}
 }
 
-func TestXAttr(t *testing.T) {
-	tc := newTestCase(t, &testOptions{attrCache: true, entryCache: true})
-
-	tc.writeOrig("file", "", 0644)
-
-	buf := make([]byte, 1024)
-	attr := "user.xattrtest"
-	if _, err := syscall.Getxattr(tc.mntDir+"/file", attr, buf); err == syscall.ENOTSUP {
-		t.Skip("$TMP does not support xattrs. Rerun this test with a $TMPDIR override")
-	}
-
-	if _, err := syscall.Getxattr(tc.mntDir+"/file", attr, buf); err != syscall.ENODATA {
-		t.Fatalf("got %v want ENOATTR", err)
-	}
-	value := []byte("value")
-	if err := syscall.Setxattr(tc.mntDir+"/file", attr, value, 0); err != nil {
-		t.Fatalf("Setxattr: %v", err)
-	}
-
-	sz, err := syscall.Listxattr(tc.mntDir+"/file", nil)
-	if err != nil {
-		t.Fatalf("Listxattr: %v", err)
-	}
-	buf = make([]byte, sz)
-	if _, err := syscall.Listxattr(tc.mntDir+"/file", buf); err != nil {
-		t.Fatalf("Listxattr: %v", err)
-	} else {
-		attributes := bytes.Split(buf[:sz], []byte{0})
-		found := false
-		for _, a := range attributes {
-			if string(a) == attr {
-				found = true
-				break
-			}
-		}
-
-		if !found {
-			t.Fatalf("Listxattr: %q (not found: %q", buf[:sz], attr)
-		}
-	}
-
-	sz, err = syscall.Getxattr(tc.mntDir+"/file", attr, buf)
-	if err != nil {
-		t.Fatalf("Getxattr: %v", err)
-	}
-	if bytes.Compare(buf[:sz], value) != 0 {
-		t.Fatalf("Getxattr got %q want %q", buf[:sz], value)
-	}
-	if err := syscall.Removexattr(tc.mntDir+"/file", attr); err != nil {
-		t.Fatalf("Removexattr: %v", err)
-	}
-
-	if _, err := syscall.Getxattr(tc.mntDir+"/file", attr, buf); err != syscall.ENODATA {
-		t.Fatalf("got %v want ENOATTR", err)
-	}
-}
-
 // TestXAttrSymlink verifies that we did not forget to use Lgetxattr instead
 // of Getxattr. This test is Linux-specific because it depends on the behavoir
 // of the `security` namespace.

fs/loopback_test.go

 package fs
 
 import (
+	"bytes"
+	"fmt"
 	"os"
 	"reflect"
 	"syscall"
 	"testing"
 
+	"github.com/hanwen/go-fuse/v2/internal/renameat"
 	"github.com/kylelemons/godebug/pretty"
 	"golang.org/x/sys/unix"
 )
@@ -38,8 +41,12 @@ func TestRenameExchange(t *testing.T) {
 		t.Fatalf("Fstatat: %v", err)
 	}
 
-	if err := unix.Renameat2(f1, "file", f2, "file", unix.RENAME_EXCHANGE); err != nil {
-		t.Errorf("rename EXCHANGE: %v", err)
+	if err := renameat.Renameat(f1, "file", f2, "file", renameat.RENAME_EXCHANGE); err != nil {
+		if err == unix.ENOSYS {
+			t.Skip("rename EXCHANGE not support on current system")
+		} else {
+			t.Errorf("rename EXCHANGE: %v", err)
+		}
 	}
 
 	var after1, after2 unix.Stat_t
@@ -81,3 +88,62 @@ func TestRenameExchange(t *testing.T) {
 		t.Errorf("got inode %d for %q want %d", ino2.StableAttr().Ino, "dir/file", after2.Ino)
 	}
 }
+
+func TestXAttr(t *testing.T) {
+	tc := newTestCase(t, &testOptions{attrCache: true, entryCache: true})
+
+	tc.writeOrig("file", "", 0644)
+
+	buf := make([]byte, 1024)
+	attrNameSpace := "user"
+	attrName := "xattrtest"
+	attr := fmt.Sprintf("%s.%s", attrNameSpace, attrName)
+	if _, err := unix.Getxattr(tc.mntDir+"/file", attr, buf); err == unix.ENOTSUP {
+		t.Skip("$TMP does not support xattrs. Rerun this test with a $TMPDIR override")
+	}
+
+	if _, err := unix.Getxattr(tc.mntDir+"/file", attr, buf); err != ENOATTR {
+		t.Fatalf("got %v want ENOATTR", err)
+	}
+	value := []byte("value")
+	if err := unix.Setxattr(tc.mntDir+"/file", attr, value, 0); err != nil {
+		t.Fatalf("Setxattr: %v", err)
+	}
+
+	sz, err := unix.Listxattr(tc.mntDir+"/file", nil)
+	if err != nil {
+		t.Fatalf("Listxattr: %v", err)
+	}
+	buf = make([]byte, sz)
+	if _, err := unix.Listxattr(tc.mntDir+"/file", buf); err != nil {
+		t.Fatalf("Listxattr: %v", err)
+	} else {
+		attributes := retrieveAttrName(buf[:sz])
+		found := false
+		for _, a := range attributes {
+			if string(a) == attr || attrNameSpace+string(a) == attr {
+				found = true
+				break
+			}
+		}
+
+		if !found {
+			t.Fatalf("Listxattr: %q (not found: %q", attributes, attr)
+		}
+	}
+
+	sz, err = unix.Getxattr(tc.mntDir+"/file", attr, buf)
+	if err != nil {
+		t.Fatalf("Getxattr: %v", err)
+	}
+	if bytes.Compare(buf[:sz], value) != 0 {
+		t.Fatalf("Getxattr got %q want %q", buf[:sz], value)
+	}
+	if err := unix.Removexattr(tc.mntDir+"/file", attr); err != nil {
+		t.Fatalf("Removexattr: %v", err)
+	}
+
+	if _, err := unix.Getxattr(tc.mntDir+"/file", attr, buf); err != ENOATTR {
+		t.Fatalf("got %v want ENOATTR", err)
+	}
+}

fs/loopback_unix.go

+//go:build !freebsd
+
+package fs
+
+import (
+	"bytes"
+	"context"
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+func retrieveAttrName(buf []byte) [][]byte {
+	attributes := bytes.Split(buf, []byte{0})
+	return attributes
+}
+
+var _ = (NodeListxattrer)((*LoopbackNode)(nil))
+
+func (n *LoopbackNode) Listxattr(ctx context.Context, dest []byte) (uint32, syscall.Errno) {
+	sz, err := unix.Llistxattr(n.path(), dest)
+	return uint32(sz), ToErrno(err)
+}

fs/maxwrite_test.go

@@ -244,7 +244,7 @@ func bdiReadahead(mnt string) int {
 	if err != nil {
 		panic(err)
 	}
-	path := fmt.Sprintf("/sys/class/bdi/%d:%d/read_ahead_kb", unix.Major(st.Dev), unix.Minor(st.Dev))
+	path := fmt.Sprintf("/sys/class/bdi/%d:%d/read_ahead_kb", unix.Major(uint64(st.Dev)), unix.Minor(uint64(st.Dev)))
 	buf, err := ioutil.ReadFile(path)
 	if err != nil {
 		panic(err)

fs/mem_test.go

@@ -137,8 +137,7 @@ func TestDataFile(t *testing.T) {
 	if err := syscall.Lstat(mntDir+"/file", &st); err != nil {
 		t.Fatalf("Lstat: %v", err)
 	}
-
-	if want := uint32(syscall.S_IFREG | 0464); st.Mode != want {
+	if want := uint(syscall.S_IFREG | 0464); uint(st.Mode) != want {
 		t.Errorf("got mode %o, want %o", st.Mode, want)
 	}
 

fs/passthrough_test.go

+// Copyright 2024 the Go-FUSE Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package fs
+
+import (
+	"context"
+	"io"
+	"os"
+	"sync"
+	"syscall"
+	"testing"
+
+	"github.com/hanwen/go-fuse/v2/fuse"
+	"github.com/hanwen/go-fuse/v2/internal/testutil"
+)
+
+type rwRegisteringNode struct {
+	LoopbackNode
+
+	mu     sync.Mutex
+	reads  int
+	writes int
+}
+
+func (n *rwRegisteringNode) Read(ctx context.Context, f FileHandle, dest []byte, off int64) (fuse.ReadResult, syscall.Errno) {
+	n.mu.Lock()
+	defer n.mu.Unlock()
+	n.reads++
+	return f.(FileReader).Read(ctx, dest, off)
+}
+
+func (n *rwRegisteringNode) Write(ctx context.Context, f FileHandle, data []byte, off int64) (written uint32, errno syscall.Errno) {
+	n.mu.Lock()
+	defer n.mu.Unlock()
+	n.writes++
+	return f.(FileWriter).Write(ctx, data, off)
+}
+
+func TestPassthrough(t *testing.T) {
+	if os.Geteuid() != 0 {
+		t.Skip("passthrough requires CAP_SYS_ADMIN")
+	}
+
+	mnt := t.TempDir()
+	n := &rwRegisteringNode{}
+
+	rootData := &LoopbackRoot{
+		Path: t.TempDir(),
+		NewNode: func(rootData *LoopbackRoot, parent *Inode, name string, st *syscall.Stat_t) InodeEmbedder {
+			return n
+		},
+	}
+	n.RootData = rootData
+	root := &LoopbackNode{
+		RootData: rootData,
+	}
+	opts := &Options{}
+	opts.Debug = testutil.VerboseTest()
+	server, err := Mount(mnt, root, opts)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer server.Unmount()
+
+	if 0 == server.KernelSettings().Flags64()&fuse.CAP_PASSTHROUGH {
+		t.Skip("Kernel does not support passthrough")
+	}
+	fn := mnt + "/file"
+	want := "hello there"
+	if err := os.WriteFile(fn, []byte(want), 0666); err != nil {
+		t.Fatalf("WriteFile: %v", err)
+	}
+
+	f, err := os.Open(fn)
+	if err != nil {
+		t.Fatalf("Open: %v", err)
+	}
+	defer f.Close()
+
+	got, err := io.ReadAll(f)
+	if err != nil {
+		t.Fatalf("Open: %v", err)
+	}
+	if want != string(got) {
+		t.Errorf("got %q want %q", got, want)
+	}
+
+	want2 := "xxxx"
+	if err := os.WriteFile(fn, []byte(want2), 0666); err != nil {
+		t.Fatalf("WriteFile: %v", err)
+	}
+
+	got2, err := os.ReadFile(fn)
+	if err != nil {
+		t.Fatalf("ReadFile: %v", err)
+	}
+	if string(got2) != want2 {
+		t.Errorf("got %q want %q", got2, want2)
+	}
+
+	f.Close()
+	server.Unmount()
+
+	if n.reads > 0 {
+		t.Errorf("got readcount %d want 0", n.reads)
+	}
+	if n.writes > 0 {
+		t.Errorf("got writecount %d want 0", n.writes)
+	}
+}

fs/readonly_test.go

@@ -11,14 +11,14 @@ import (
 	"testing"
 
 	"github.com/hanwen/go-fuse/v2/fuse"
+	"golang.org/x/sys/unix"
 )
 
 func TestReadonlyCreate(t *testing.T) {
 	root := &Inode{}
 
 	mntDir, _ := testMount(t, root, nil)
-
-	_, err := syscall.Creat(mntDir+"/test", 0644)
+	_, err := unix.Open(mntDir+"/test", unix.O_CREAT, 0644)
 	if want := syscall.EROFS; want != err {
 		t.Fatalf("got err %v, want %v", err, want)
 	}
@@ -44,7 +44,7 @@ func TestDefaultPermissions(t *testing.T) {
 		var st syscall.Stat_t
 		if err := syscall.Lstat(filepath.Join(mntDir, k), &st); err != nil {
 			t.Error("Lstat", err)
-		} else if st.Mode != v {
+		} else if uint(st.Mode) != uint(v) {
 			t.Errorf("got %o want %o", st.Mode, v)
 		}
 	}

fs/simple_test.go

@@ -25,6 +25,7 @@ import (
 	"github.com/hanwen/go-fuse/v2/fuse"
 	"github.com/hanwen/go-fuse/v2/internal/testutil"
 	"github.com/hanwen/go-fuse/v2/posixtest"
+	"golang.org/x/sys/unix"
 )
 
 type testCase struct {
@@ -339,7 +340,7 @@ func TestMknod(t *testing.T) {
 			var st syscall.Stat_t
 			if err := syscall.Stat(p, &st); err != nil {
 				got := st.Mode &^ 07777
-				if want := mode; got != want {
+				if want := uint(mode); want != uint(got) {
 					t.Fatalf("stat(%s): got %o want %o", nm, got, want)
 				}
 			}
@@ -646,28 +647,38 @@ func TestStaleHardlinks(t *testing.T) {
 	// Disable all caches we can disable
 	tc := newTestCase(t, &testOptions{attrCache: false, entryCache: false})
 
+	// gvfsd-trash sets an inotify watch on mntDir and stat()s every file that is
+	// created, racing with the test logic ( https://github.com/hanwen/go-fuse/issues/478 ).
+	// Use a subdir to prevent that.
+	if err := os.Mkdir(tc.mntDir+"/x", 0755); err != nil {
+		t.Fatal(err)
+	}
+
 	// "link0" is original file
-	link0 := tc.mntDir + "/link0"
-	if fd, err := syscall.Creat(link0, 0600); err != nil {
+	link0 := tc.mntDir + "/x/link0"
+	if fd, err := unix.Open(link0, unix.O_CREAT, 0600); err != nil {
 		t.Fatal(err)
 	} else {
 		syscall.Close(fd)
 	}
 	// Create hardlinks via mntDir
+	t.Logf("create link1...20, pid=%d", os.Getpid())
 	for i := 1; i < 20; i++ {
-		linki := fmt.Sprintf(tc.mntDir+"/link%d", i)
+		linki := fmt.Sprintf(tc.mntDir+"/x/link%d", i)
 		if err := syscall.Link(link0, linki); err != nil {
 			t.Fatal(err)
 		}
 	}
 	// Delete hardlinks via origDir (behind loopback fs's back)
+	t.Log("delete link1...20 behind loopback's back")
 	for i := 1; i < 20; i++ {
-		linki := fmt.Sprintf(tc.origDir+"/link%d", i)
+		linki := fmt.Sprintf(tc.origDir+"/x/link%d", i)
 		if err := syscall.Unlink(linki); err != nil {
 			t.Fatal(err)
 		}
 	}
 	// Try to open link0 via mntDir
+	t.Log("open link0")
 	fd, err := syscall.Open(link0, syscall.O_RDONLY, 0)
 	if err != nil {
 		t.Error(err)
@@ -680,3 +691,47 @@ func TestStaleHardlinks(t *testing.T) {
 func init() {
 	syscall.Umask(0)
 }
+
+func testMountDir(dir string) error {
+	opts := &Options{}
+	opts.Debug = testutil.VerboseTest()
+	server, err := Mount(dir, &Inode{}, opts)
+	if err != nil {
+		return err
+	}
+
+	server.Unmount()
+	server.Wait()
+	return nil
+}
+
+func TestParallelMount(t *testing.T) {
+	before := runtime.GOMAXPROCS(1)
+	defer runtime.GOMAXPROCS(before)
+	// Per default, only 1000 FUSE mounts are allowed, then you get
+	// > /usr/bin/fusermount3: too many FUSE filesystems mounted; mount_max=N can be set in /etc/fuse.conf
+	// Let's stay well below 1000.
+	N := 900
+	todo := make(chan string, N)
+	result := make(chan error, N)
+	for i := 0; i < N; i++ {
+		todo <- t.TempDir()
+	}
+	close(todo)
+
+	P := 2
+	for i := 0; i < P; i++ {
+		go func() {
+			for d := range todo {
+				result <- testMountDir(d)
+			}
+		}()
+	}
+
+	for i := 0; i < N; i++ {
+		e := <-result
+		if e != nil {
+			t.Error(e)
+		}
+	}
+}

fuse/api.go

@@ -289,9 +289,10 @@ type MountOptions struct {
 // 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.
+// this, but rather the interfaces associated with
+// fs.InodeEmbedder. 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

fuse/attr_darwin.go → fuse/attr_unix.go

-// Copyright 2016 the Go-FUSE Authors. All rights reserved.
+//go:build !linux
+
+// Copyright 2024 the Go-FUSE Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 

fuse/misc.go

@@ -70,11 +70,11 @@ func CurrentOwner() *Owner {
 	}
 }
 
-const _UTIME_OMIT = ((1 << 30) - 2)
-
 // 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.
+//
+// Deprecated: use unix.TimeToTimespec from the x/sys/unix package instead.
 func UtimeToTimespec(t *time.Time) (ts syscall.Timespec) {
 	if t == nil {
 		ts.Nsec = _UTIME_OMIT

fuse/misc_darwin.go

+package fuse
+
+// Ref: https://github.com/apple/darwin-xnu/blob/main/bsd/sys/stat.h#L576
+const _UTIME_OMIT = -2

fuse/misc_unix.go

+//go:build !darwin
+
+package fuse
+
+import "golang.org/x/sys/unix"
+
+const _UTIME_OMIT = unix.UTIME_OMIT

fuse/mount.go

@@ -7,6 +7,30 @@ import (
 	"syscall"
 )
 
+var reservedFDs []*os.File
+
+func init() {
+	// Both Darwin and Linux invoke a subprocess with one
+	// inherited file descriptor to create the mount. To protect
+	// against deadlock, we must ensure that file descriptor 3
+	// never points to a FUSE filesystem. We do this by simply
+	// grabbing fd 3 and never releasing it. (This is not
+	// completely foolproof: a preceding init routine could grab fd 3,
+	// and then release it later.)
+	for {
+		r, w, err := os.Pipe()
+		if err != nil {
+			panic(fmt.Sprintf("os.Pipe(): %v", err))
+		}
+		w.Close()
+		if r.Fd() > 3 {
+			r.Close()
+			break
+		}
+		reservedFDs = append(reservedFDs, r)
+	}
+}
+
 func getConnection(local *os.File) (int, error) {
 	conn, err := net.FileConn(local)
 	if err != nil {

fuse/mount_freebsd.go

+package fuse
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+	"syscall"
+)
+
+func callMountFuseFs(mountPoint string, opts *MountOptions) (fd int, err error) {
+	bin, err := fusermountBinary()
+	if err != nil {
+		return 0, err
+	}
+	f, err := os.OpenFile("/dev/fuse", os.O_RDWR, 0o000)
+	if err != nil {
+		return -1, err
+	}
+	cmd := exec.Command(
+		bin,
+		"--safe",
+		"-o", strings.Join(opts.optionsStrings(), ","),
+		"3",
+		mountPoint,
+	)
+	cmd.Env = []string{"MOUNT_FUSEFS_CALL_BY_LIB=1"}
+	cmd.ExtraFiles = []*os.File{f}
+
+	if err := cmd.Start(); err != nil {
+		f.Close()
+		return -1, fmt.Errorf("mount_fusefs: %v", err)
+	}
+
+	if err := cmd.Wait(); err != nil {
+		// see if we have a better error to report
+		f.Close()
+		return -1, fmt.Errorf("mount_fusefs: %v", err)
+	}
+
+	return int(f.Fd()), nil
+}
+
+func mount(mountPoint string, opts *MountOptions, ready chan<- error) (fd int, err error) {
+	// Using the same logic from libfuse to prevent chaos
+	for {
+		f, err := os.OpenFile("/dev/null", os.O_RDWR, 0o000)
+		if err != nil {
+			return -1, err
+		}
+		if f.Fd() > 2 {
+			f.Close()
+			break
+		}
+	}
+
+	// Magic `/dev/fd/N` mountpoint. See the docs for NewServer() for how this
+	// works.
+
+	fd = parseFuseFd(mountPoint)
+	if fd >= 0 {
+		if opts.Debug {
+			opts.Logger.Printf("mount: magic mountpoint %q, using fd %d", mountPoint, fd)
+		}
+	} else {
+		// Usual case: mount via the `fusermount` suid helper
+		fd, err = callMountFuseFs(mountPoint, opts)
+		if err != nil {
+			return
+		}
+	}
+	// golang sets CLOEXEC on file descriptors when they are
+	// acquired through normal operations (e.g. open).
+	// Buf for fd, we have to set CLOEXEC manually
+	syscall.CloseOnExec(fd)
+
+	close(ready)
+	return fd, err
+}
+
+func unmount(mountPoint string, opts *MountOptions) (err error) {
+	_ = opts
+	return syscall.Unmount(mountPoint, 0)
+}
+
+func fusermountBinary() (string, error) {
+	binPaths := []string{
+		"/sbin/mount_fusefs",
+	}
+
+	for _, path := range binPaths {
+		if _, err := os.Stat(path); err == nil {
+			return path, nil
+		}
+	}
+
+	return "", fmt.Errorf("no FUSE mount utility found")
+}

fuse/opcode.go

@@ -99,40 +99,41 @@ func doInit(server *Server, req *request) {
 		return
 	}
 
+	kernelFlags := input.Flags64()
 	server.reqMu.Lock()
 	server.kernelSettings = *input
-	server.kernelSettings.Flags = input.Flags & (CAP_ASYNC_READ | CAP_BIG_WRITES | CAP_FILE_OPS |
-		CAP_READDIRPLUS | CAP_NO_OPEN_SUPPORT | CAP_PARALLEL_DIROPS | CAP_MAX_PAGES | CAP_RENAME_SWAP)
+	kernelFlags &= (CAP_ASYNC_READ | CAP_BIG_WRITES | CAP_FILE_OPS |
+		CAP_READDIRPLUS | CAP_NO_OPEN_SUPPORT | CAP_PARALLEL_DIROPS | CAP_MAX_PAGES | CAP_RENAME_SWAP | CAP_PASSTHROUGH)
 
 	if server.opts.EnableLocks {
-		server.kernelSettings.Flags |= CAP_FLOCK_LOCKS | CAP_POSIX_LOCKS
+		kernelFlags |= CAP_FLOCK_LOCKS | CAP_POSIX_LOCKS
 	}
 	if server.opts.EnableSymlinkCaching {
-		server.kernelSettings.Flags |= CAP_CACHE_SYMLINKS
+		kernelFlags |= CAP_CACHE_SYMLINKS
 	}
 	if server.opts.EnableAcl {
-		server.kernelSettings.Flags |= CAP_POSIX_ACL
+		kernelFlags |= CAP_POSIX_ACL
 	}
 	if server.opts.SyncRead {
 		// Clear CAP_ASYNC_READ
-		server.kernelSettings.Flags &= ^uint32(CAP_ASYNC_READ)
+		kernelFlags &= ^uint64(CAP_ASYNC_READ)
 	}
 	if server.opts.DisableReadDirPlus {
 		// Clear CAP_READDIRPLUS
-		server.kernelSettings.Flags &= ^uint32(CAP_READDIRPLUS)
+		kernelFlags &= ^uint64(CAP_READDIRPLUS)
 	}
 
-	dataCacheMode := input.Flags & CAP_AUTO_INVAL_DATA
+	dataCacheMode := kernelFlags & CAP_AUTO_INVAL_DATA
 	if server.opts.ExplicitDataCacheControl {
 		// we don't want CAP_AUTO_INVAL_DATA even if we cannot go into fully explicit mode
 		dataCacheMode = 0
 
-		explicit := input.Flags & CAP_EXPLICIT_INVAL_DATA
+		explicit := kernelFlags & CAP_EXPLICIT_INVAL_DATA
 		if explicit != 0 {
 			dataCacheMode = explicit
 		}
 	}
-	server.kernelSettings.Flags |= dataCacheMode
+	kernelFlags |= dataCacheMode
 
 	if input.Minor >= 13 {
 		server.setSplice()
@@ -149,13 +150,13 @@ func doInit(server *Server, req *request) {
 		Major:               _FUSE_KERNEL_VERSION,
 		Minor:               _OUR_MINOR_VERSION,
 		MaxReadAhead:        input.MaxReadAhead,
-		Flags:               server.kernelSettings.Flags,
 		MaxWrite:            uint32(server.opts.MaxWrite),
 		CongestionThreshold: uint16(server.opts.MaxBackground * 3 / 4),
 		MaxBackground:       uint16(server.opts.MaxBackground),
 		MaxPages:            uint16(maxPages),
+		MaxStackDepth:       1,
 	}
-
+	out.setFlags(kernelFlags)
 	if server.opts.MaxReadAhead != 0 && uint32(server.opts.MaxReadAhead) < out.MaxReadAhead {
 		out.MaxReadAhead = uint32(server.opts.MaxReadAhead)
 	}

fuse/passthrough_linux.go

+// Copyright 2024 the Go-FUSE Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package fuse
+
+import (
+	"syscall"
+	"unsafe"
+)
+
+const (
+	_DEV_IOC_BACKING_OPEN  = 0x4010e501
+	_DEV_IOC_BACKING_CLOSE = 0x4004e502
+)
+
+// RegisterBackingFd registers the given file descriptor in the
+// kernel, so the kernel can bypass FUSE and access the backing file
+// directly for read and write calls. On success a backing ID is
+// returned. The backing ID should unregistered using
+// UnregisterBackingFd() once the file is released.  Within the
+// kernel, an inode can only have a single backing file, so multiple
+// Open/Create calls should coordinate to return a consistent backing
+// ID.
+func (ms *Server) RegisterBackingFd(m *BackingMap) (int32, syscall.Errno) {
+	ms.writeMu.Lock()
+	id, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(ms.mountFd), uintptr(_DEV_IOC_BACKING_OPEN), uintptr(unsafe.Pointer(m)))
+	ms.writeMu.Unlock()
+	if ms.opts.Debug {
+		ms.opts.Logger.Printf("ioctl: BACKING_OPEN %v: id %d (%v)", m.string(), id, errno)
+	}
+	return int32(id), errno
+}
+
+// UnregisterBackingFd unregisters the given ID in the kernel. The ID
+// should have been acquired before using RegisterBackingFd.
+func (ms *Server) UnregisterBackingFd(id int32) syscall.Errno {
+	ms.writeMu.Lock()
+	_, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(ms.mountFd), uintptr(_DEV_IOC_BACKING_CLOSE), uintptr(unsafe.Pointer(&id)))
+	ms.writeMu.Unlock()
+
+	if ms.opts.Debug {
+		ms.opts.Logger.Printf("ioctl: BACKING_CLOSE id %d: %v", id, errno)
+	}
+	return errno
+}

fuse/poll_linux.go → fuse/poll_unix.go

+//go:build !darwin
+
 package fuse
 
 import (

fuse/print.go

@@ -12,76 +12,82 @@ import (
 )
 
 var (
-	writeFlagNames = newFlagNames(map[int64]string{
-		WRITE_CACHE:     "CACHE",
-		WRITE_LOCKOWNER: "LOCKOWNER",
+	isTest         bool
+	writeFlagNames = newFlagNames([]flagNameEntry{
+		{WRITE_CACHE, "CACHE"},
+		{WRITE_LOCKOWNER, "LOCKOWNER"},
 	})
-	readFlagNames = newFlagNames(map[int64]string{
-		READ_LOCKOWNER: "LOCKOWNER",
+	readFlagNames = newFlagNames([]flagNameEntry{
+		{READ_LOCKOWNER, "LOCKOWNER"},
 	})
-	initFlagNames = newFlagNames(map[int64]string{
-		CAP_ASYNC_READ:        "ASYNC_READ",
-		CAP_POSIX_LOCKS:       "POSIX_LOCKS",
-		CAP_FILE_OPS:          "FILE_OPS",
-		CAP_ATOMIC_O_TRUNC:    "ATOMIC_O_TRUNC",
-		CAP_EXPORT_SUPPORT:    "EXPORT_SUPPORT",
-		CAP_BIG_WRITES:        "BIG_WRITES",
-		CAP_DONT_MASK:         "DONT_MASK",
-		CAP_SPLICE_WRITE:      "SPLICE_WRITE",
-		CAP_SPLICE_MOVE:       "SPLICE_MOVE",
-		CAP_SPLICE_READ:       "SPLICE_READ",
-		CAP_FLOCK_LOCKS:       "FLOCK_LOCKS",
-		CAP_IOCTL_DIR:         "IOCTL_DIR",
-		CAP_AUTO_INVAL_DATA:   "AUTO_INVAL_DATA",
-		CAP_READDIRPLUS:       "READDIRPLUS",
-		CAP_READDIRPLUS_AUTO:  "READDIRPLUS_AUTO",
-		CAP_ASYNC_DIO:         "ASYNC_DIO",
-		CAP_WRITEBACK_CACHE:   "WRITEBACK_CACHE",
-		CAP_NO_OPEN_SUPPORT:   "NO_OPEN_SUPPORT",
-		CAP_PARALLEL_DIROPS:   "PARALLEL_DIROPS",
-		CAP_POSIX_ACL:         "POSIX_ACL",
-		CAP_HANDLE_KILLPRIV:   "HANDLE_KILLPRIV",
-		CAP_ABORT_ERROR:       "ABORT_ERROR",
-		CAP_MAX_PAGES:         "MAX_PAGES",
-		CAP_CACHE_SYMLINKS:    "CACHE_SYMLINKS",
-		CAP_SECURITY_CTX:      "SECURITY_CTX",
-		CAP_HAS_INODE_DAX:     "HAS_INODE_DAX",
-		CAP_CREATE_SUPP_GROUP: "CREATE_SUPP_GROUP",
-		CAP_HAS_EXPIRE_ONLY:   "HAS_EXPIRE_ONLY",
-		CAP_DIRECT_IO_RELAX:   "DIRECT_IO_RELAX",
+	initFlagNames = newFlagNames([]flagNameEntry{
+		{CAP_ASYNC_READ, "ASYNC_READ"},
+		{CAP_POSIX_LOCKS, "POSIX_LOCKS"},
+		{CAP_FILE_OPS, "FILE_OPS"},
+		{CAP_ATOMIC_O_TRUNC, "ATOMIC_O_TRUNC"},
+		{CAP_EXPORT_SUPPORT, "EXPORT_SUPPORT"},
+		{CAP_BIG_WRITES, "BIG_WRITES"},
+		{CAP_DONT_MASK, "DONT_MASK"},
+		{CAP_SPLICE_WRITE, "SPLICE_WRITE"},
+		{CAP_SPLICE_MOVE, "SPLICE_MOVE"},
+		{CAP_SPLICE_READ, "SPLICE_READ"},
+		{CAP_FLOCK_LOCKS, "FLOCK_LOCKS"},
+		{CAP_IOCTL_DIR, "IOCTL_DIR"},
+		{CAP_AUTO_INVAL_DATA, "AUTO_INVAL_DATA"},
+		{CAP_READDIRPLUS, "READDIRPLUS"},
+		{CAP_READDIRPLUS_AUTO, "READDIRPLUS_AUTO"},
+		{CAP_ASYNC_DIO, "ASYNC_DIO"},
+		{CAP_WRITEBACK_CACHE, "WRITEBACK_CACHE"},
+		{CAP_NO_OPEN_SUPPORT, "NO_OPEN_SUPPORT"},
+		{CAP_PARALLEL_DIROPS, "PARALLEL_DIROPS"},
+		{CAP_POSIX_ACL, "POSIX_ACL"},
+		{CAP_HANDLE_KILLPRIV, "HANDLE_KILLPRIV"},
+		{CAP_ABORT_ERROR, "ABORT_ERROR"},
+		{CAP_MAX_PAGES, "MAX_PAGES"},
+		{CAP_CACHE_SYMLINKS, "CACHE_SYMLINKS"},
+		{CAP_SECURITY_CTX, "SECURITY_CTX"},
+		{CAP_HAS_INODE_DAX, "HAS_INODE_DAX"},
+		{CAP_CREATE_SUPP_GROUP, "CREATE_SUPP_GROUP"},
+		{CAP_HAS_EXPIRE_ONLY, "HAS_EXPIRE_ONLY"},
+		{CAP_DIRECT_IO_ALLOW_MMAP, "DIRECT_IO_ALLOW_MMAP"},
+		{CAP_PASSTHROUGH, "PASSTHROUGH"},
+		{CAP_NO_EXPORT_SUPPORT, "NO_EXPORT_SUPPORT"},
+		{CAP_HAS_RESEND, "HAS_RESEND"},
 	})
-	releaseFlagNames = newFlagNames(map[int64]string{
-		RELEASE_FLUSH: "FLUSH",
+	releaseFlagNames = newFlagNames([]flagNameEntry{
+		{RELEASE_FLUSH, "FLUSH"},
 	})
-	openFlagNames = newFlagNames(map[int64]string{
-		int64(os.O_WRONLY):         "WRONLY",
-		int64(os.O_RDWR):           "RDWR",
-		int64(os.O_APPEND):         "APPEND",
-		int64(syscall.O_ASYNC):     "ASYNC",
-		int64(os.O_CREATE):         "CREAT",
-		int64(os.O_EXCL):           "EXCL",
-		int64(syscall.O_NOCTTY):    "NOCTTY",
-		int64(syscall.O_NONBLOCK):  "NONBLOCK",
-		int64(os.O_SYNC):           "SYNC",
-		int64(os.O_TRUNC):          "TRUNC",
-		0x8000:                     "LARGEFILE",
-		int64(syscall.O_CLOEXEC):   "CLOEXEC",
-		int64(syscall.O_DIRECTORY): "DIRECTORY",
+	openFlagNames = newFlagNames([]flagNameEntry{
+		{int64(os.O_WRONLY), "WRONLY"},
+		{int64(os.O_RDWR), "RDWR"},
+		{int64(os.O_APPEND), "APPEND"},
+		{int64(syscall.O_ASYNC), "ASYNC"},
+		{int64(os.O_CREATE), "CREAT"},
+		{int64(os.O_EXCL), "EXCL"},
+		{int64(syscall.O_NOCTTY), "NOCTTY"},
+		{int64(syscall.O_NONBLOCK), "NONBLOCK"},
+		{int64(os.O_SYNC), "SYNC"},
+		{int64(os.O_TRUNC), "TRUNC"},
+		{int64(syscall.O_CLOEXEC), "CLOEXEC"},
+		{int64(syscall.O_DIRECTORY), "DIRECTORY"},
 	})
-	fuseOpenFlagNames = newFlagNames(map[int64]string{
-		FOPEN_DIRECT_IO:   "DIRECT",
-		FOPEN_KEEP_CACHE:  "CACHE",
-		FOPEN_NONSEEKABLE: "NONSEEK",
-		FOPEN_CACHE_DIR:   "CACHE_DIR",
-		FOPEN_STREAM:      "STREAM",
+	fuseOpenFlagNames = newFlagNames([]flagNameEntry{
+		{FOPEN_DIRECT_IO, "DIRECT"},
+		{FOPEN_KEEP_CACHE, "CACHE"},
+		{FOPEN_NONSEEKABLE, "NONSEEK"},
+		{FOPEN_CACHE_DIR, "CACHE_DIR"},
+		{FOPEN_STREAM, "STREAM"},
+		{FOPEN_NOFLUSH, "NOFLUSH"},
+		{FOPEN_PARALLEL_DIRECT_WRITES, "PARALLEL_DIRECT_WRITES"},
+		{FOPEN_PASSTHROUGH, "PASSTHROUGH"},
 	})
-	accessFlagName = newFlagNames(map[int64]string{
-		X_OK: "x",
-		W_OK: "w",
-		R_OK: "r",
+	accessFlagName = newFlagNames([]flagNameEntry{
+		{X_OK, "x"},
+		{W_OK, "w"},
+		{R_OK, "r"},
 	})
-	getAttrFlagNames = newFlagNames(map[int64]string{
-		FUSE_GETATTR_FH: "FH",
+	getAttrFlagNames = newFlagNames([]flagNameEntry{
+		{FUSE_GETATTR_FH, "FH"},
 	})
 )
 
@@ -99,10 +105,10 @@ type flagNameEntry struct {
 }
 
 // newFlagNames creates flagNames from flag->name map.
-func newFlagNames(names map[int64]string) *flagNames {
+func newFlagNames(names []flagNameEntry) *flagNames {
 	var v flagNames
-	for flag, name := range names {
-		v.set(flag, name)
+	for _, e := range names {
+		v.set(e.bits, e.name)
 	}
 	return &v
 }
@@ -112,7 +118,7 @@ func (names *flagNames) set(flag int64, name string) {
 	entry := flagNameEntry{bits: flag, name: name}
 	for i := 0; i < 64; i++ {
 		if flag&(1<<i) != 0 {
-			if ie := names[i]; ie.bits != 0 {
+			if ie := names[i]; ie.bits != 0 && isTest {
 				panic(fmt.Sprintf("%s (%x) overlaps with %s (%x)", name, flag, ie.name, ie.bits))
 			}
 			names[i] = entry
@@ -205,22 +211,26 @@ func (in *OpenIn) string() string {
 }
 
 func (in *OpenOut) string() string {
-	return fmt.Sprintf("{Fh %d %s}", in.Fh,
+	backing := ""
+	if in.BackingID != 0 {
+		backing = fmt.Sprintf("backing=%d ", in.BackingID)
+	}
+	return fmt.Sprintf("{Fh %d %s%s}", in.Fh, backing,
 		flagString(fuseOpenFlagNames, int64(in.OpenFlags), ""))
 }
 
 func (in *InitIn) string() string {
 	return fmt.Sprintf("{%d.%d Ra %d %s}",
 		in.Major, in.Minor, in.MaxReadAhead,
-		flagString(initFlagNames, int64(in.Flags)|(int64(in.Flags2)<<32), ""))
+		flagString(initFlagNames, int64(in.Flags64()), ""))
 }
 
 func (o *InitOut) string() string {
-	return fmt.Sprintf("{%d.%d Ra %d %s %d/%d Wr %d Tg %d MaxPages %d}",
+	return fmt.Sprintf("{%d.%d Ra %d %s %d/%d Wr %d Tg %d MaxPages %d MaxStack %d}",
 		o.Major, o.Minor, o.MaxReadAhead,
-		flagString(initFlagNames, int64(o.Flags), ""),
+		flagString(initFlagNames, int64(o.Flags64()), ""),
 		o.CongestionThreshold, o.MaxBackground, o.MaxWrite,
-		o.TimeGran, o.MaxPages)
+		o.TimeGran, o.MaxPages, o.MaxStackDepth)
 }
 
 func (s *FsyncIn) string() string {
@@ -269,7 +279,7 @@ func (o *EntryOut) string() string {
 }
 
 func (o *CreateOut) string() string {
-	return fmt.Sprintf("{n%d g%d %v %v}", o.NodeId, o.Generation, &o.EntryOut, &o.OpenOut)
+	return fmt.Sprintf("{n%d g%d %v %v}", o.NodeId, o.Generation, &o.EntryOut, o.OpenOut.string())
 }
 
 func (o *StatfsOut) string() string {
@@ -352,3 +362,22 @@ func Print(obj interface{}) string {
 	}
 	return fmt.Sprintf("%T: %v", obj, obj)
 }
+
+func (a *Attr) string() string {
+	return fmt.Sprintf(
+		"{M0%o SZ=%d L=%d "+
+			"%d:%d "+
+			"B%d*%d i%d:%d "+
+			"A %f "+
+			"M %f "+
+			"C %f}",
+		a.Mode, a.Size, a.Nlink,
+		a.Uid, a.Gid,
+		a.Blocks, a.Blksize,
+		a.Rdev, a.Ino, ft(a.Atime, a.Atimensec), ft(a.Mtime, a.Mtimensec),
+		ft(a.Ctime, a.Ctimensec))
+}
+
+func (m *BackingMap) string() string {
+	return fmt.Sprintf("{fd %d, flags 0x%x}", m.Fd, m.Flags)
+}

fuse/print_darwin.go

@@ -19,21 +19,6 @@ func init() {
 	initFlagNames.set(CAP_CASE_INSENSITIVE, "CASE_INSENSITIVE")
 }
 
-func (a *Attr) string() string {
-	return fmt.Sprintf(
-		"{M0%o SZ=%d L=%d "+
-			"%d:%d "+
-			"B%d*%d i%d:%d "+
-			"A %f "+
-			"M %f "+
-			"C %f}",
-		a.Mode, a.Size, a.Nlink,
-		a.Uid, a.Gid,
-		a.Blocks, a.Blksize,
-		a.Rdev, a.Ino, ft(a.Atime, a.Atimensec), ft(a.Mtime, a.Mtimensec),
-		ft(a.Ctime, a.Ctimensec))
-}
-
 func (me *CreateIn) string() string {
 	return fmt.Sprintf(
 		"{0%o [%s]}", me.Mode,

fuse/print_freebsd.go

+package fuse
+
+func init() {
+	initFlagNames.set(CAP_NO_OPENDIR_SUPPORT, "NO_OPENDIR_SUPPORT")
+}

fuse/print_linux.go

@@ -5,13 +5,20 @@
 package fuse
 
 import (
-	"fmt"
+	"runtime"
+	"strings"
 	"syscall"
 )
 
 func init() {
+	// syscall.O_LARGEFILE is 0x0 on x86_64, but the kernel
+	// supplies 0x8000 anyway, except on mips64el, where 0x8000 is
+	// used for O_DIRECT.
+	if !strings.Contains(runtime.GOARCH, "mips64") {
+		openFlagNames.set(0x8000, "LARGEFILE")
+	}
+
 	openFlagNames.set(syscall.O_DIRECT, "DIRECT")
-	openFlagNames.set(syscall.O_LARGEFILE, "LARGEFILE")
 	openFlagNames.set(syscall_O_NOATIME, "NOATIME")
 	initFlagNames.set(CAP_NO_OPENDIR_SUPPORT, "NO_OPENDIR_SUPPORT")
 	initFlagNames.set(CAP_EXPLICIT_INVAL_DATA, "EXPLICIT_INVAL_DATA")
@@ -22,48 +29,3 @@ func init() {
 	initFlagNames.set(CAP_INIT_EXT, "INIT_EXT")
 	initFlagNames.set(CAP_INIT_RESERVED, "INIT_RESERVED")
 }
-
-func (a *Attr) string() string {
-	return fmt.Sprintf(
-		"{M0%o SZ=%d L=%d "+
-			"%d:%d "+
-			"B%d*%d i%d:%d "+
-			"A %f "+
-			"M %f "+
-			"C %f}",
-		a.Mode, a.Size, a.Nlink,
-		a.Uid, a.Gid,
-		a.Blocks, a.Blksize,
-		a.Rdev, a.Ino, ft(a.Atime, a.Atimensec), ft(a.Mtime, a.Mtimensec),
-		ft(a.Ctime, a.Ctimensec))
-}
-
-func (in *CreateIn) string() string {
-	return fmt.Sprintf(
-		"{0%o [%s] (0%o)}", in.Mode,
-		flagString(openFlagNames, int64(in.Flags), "O_RDONLY"), in.Umask)
-}
-
-func (in *GetAttrIn) string() string {
-	return fmt.Sprintf("{Fh %d %s}", in.Fh_, flagString(getAttrFlagNames, int64(in.Flags_), ""))
-}
-
-func (in *MknodIn) string() string {
-	return fmt.Sprintf("{0%o (0%o), %d}", in.Mode, in.Umask, in.Rdev)
-}
-
-func (in *ReadIn) string() string {
-	return fmt.Sprintf("{Fh %d [%d +%d) %s L %d %s}",
-		in.Fh, in.Offset, in.Size,
-		flagString(readFlagNames, int64(in.ReadFlags), ""),
-		in.LockOwner,
-		flagString(openFlagNames, int64(in.Flags), "RDONLY"))
-}
-
-func (in *WriteIn) string() string {
-	return fmt.Sprintf("{Fh %d [%d +%d) %s L %d %s}",
-		in.Fh, in.Offset, in.Size,
-		flagString(writeFlagNames, int64(in.WriteFlags), ""),
-		in.LockOwner,
-		flagString(openFlagNames, int64(in.Flags), "RDONLY"))
-}

fuse/print_test.go

@@ -8,6 +8,10 @@ import (
 	"testing"
 )
 
+func init() {
+	isTest = true
+}
+
 // verify that flagString always formats flags in the same order.
 func TestFlagStringOrder(t *testing.T) {
 	var flags int64 = CAP_ASYNC_READ | CAP_SPLICE_WRITE | CAP_READDIRPLUS | CAP_MAX_PAGES | CAP_EXPLICIT_INVAL_DATA
@@ -23,10 +27,10 @@ func TestFlagStringOrder(t *testing.T) {
 
 // verify how flagString handles provided default.
 func TestFlagStringDefault(t *testing.T) {
-	names := newFlagNames(map[int64]string{
-		1: "AAA",
-		2: "BBB",
-		4: "CCC",
+	names := newFlagNames([]flagNameEntry{
+		{1, "AAA"},
+		{2, "BBB"},
+		{4, "CCC"},
 	})
 
 	testv := []struct {

fuse/print_unix.go

+//go:build !darwin
+
+package fuse
+
+import "fmt"
+
+func (in *CreateIn) string() string {
+	return fmt.Sprintf(
+		"{0%o [%s] (0%o)}", in.Mode,
+		flagString(openFlagNames, int64(in.Flags), "O_RDONLY"), in.Umask)
+}
+
+func (in *GetAttrIn) string() string {
+	return fmt.Sprintf("{Fh %d %s}", in.Fh_, flagString(getAttrFlagNames, int64(in.Flags_), ""))
+}
+
+func (in *MknodIn) string() string {
+	return fmt.Sprintf("{0%o (0%o), %d}", in.Mode, in.Umask, in.Rdev)
+}
+
+func (in *ReadIn) string() string {
+	return fmt.Sprintf("{Fh %d [%d +%d) %s L %d %s}",
+		in.Fh, in.Offset, in.Size,
+		flagString(readFlagNames, int64(in.ReadFlags), ""),
+		in.LockOwner,
+		flagString(openFlagNames, int64(in.Flags), "RDONLY"))
+}
+
+func (in *WriteIn) string() string {
+	return fmt.Sprintf("{Fh %d [%d +%d) %s L %d %s}",
+		in.Fh, in.Offset, in.Size,
+		flagString(writeFlagNames, int64(in.WriteFlags), ""),
+		in.LockOwner,
+		flagString(openFlagNames, int64(in.Flags), "RDONLY"))
+}

fuse/request.go

@@ -103,9 +103,9 @@ func (r *request) InputDebug() string {
 		names += fmt.Sprintf("%s %db", data, len(r.arg))
 	}
 
-	return fmt.Sprintf("rx %d: %s n%d %s%s",
+	return fmt.Sprintf("rx %d: %s n%d %s%s p%d",
 		r.inHeader.Unique, operationName(r.inHeader.Opcode), r.inHeader.NodeId,
-		val, names)
+		val, names, r.inHeader.Caller.Pid)
 }
 
 func (r *request) OutputDebug() string {

fuse/request_freebsd.go

+package fuse
+
+const outputHeaderSize = 160
+
+const (
+	_FUSE_KERNEL_VERSION   = 7
+	_MINIMUM_MINOR_VERSION = 12
+	_OUR_MINOR_VERSION     = 28
+)

fuse/server.go

@@ -76,6 +76,7 @@ type Server struct {
 	singleReader bool
 	canSplice    bool
 	loops        sync.WaitGroup
+	serving      bool // for preventing duplicate Serve() calls
 
 	ready chan error
 
@@ -195,14 +196,11 @@ func NewServer(fs RawFileSystem, mountPoint string, opts *MountOptions) (*Server
 	}
 
 	ms := &Server{
-		fileSystem:  fs,
-		opts:        &o,
-		maxReaders:  maxReaders,
-		retrieveTab: make(map[uint64]*retrieveCacheRequest),
-		// OSX has races when multiple routines read from the
-		// FUSE device: on unmount, sometime some reads do not
-		// error-out, meaning that unmount will hang.
-		singleReader: runtime.GOOS == "darwin",
+		fileSystem:   fs,
+		opts:         &o,
+		maxReaders:   maxReaders,
+		retrieveTab:  make(map[uint64]*retrieveCacheRequest),
+		singleReader: useSingleReader,
 		ready:        make(chan error, 1),
 	}
 	ms.reqPool.New = func() interface{} {
@@ -419,6 +417,14 @@ func (ms *Server) recordStats(req *request) {
 //
 // Each filesystem operation executes in a separate goroutine.
 func (ms *Server) Serve() {
+	if ms.serving {
+		// Calling Serve() multiple times leads to a panic on unmount and fun
+		// debugging sessions ( https://github.com/hanwen/go-fuse/issues/512 ).
+		// Catch it early.
+		log.Panic("Serve() must only be called once, you have called it a second time")
+	}
+	ms.serving = true
+
 	ms.loop(false)
 	ms.loops.Wait()
 
@@ -471,6 +477,33 @@ func (ms *Server) handleInit() Status {
 	return OK
 }
 
+// loop is the FUSE event loop. The simplistic way of calling this is
+// with singleReader=true, which has a single goroutine reading the
+// device, and spawning a new goroutine for each request. It is
+// however 2x slower than processing the request inline with the
+// reader. The latter requires more logic, because whenever we start
+// processing the request, we have to make sure a new routine is
+// spawned to read the device.
+//
+// Benchmark results i5-8350 pinned at 2Ghz:
+//
+// singleReader = true
+//
+// BenchmarkGoFuseRead            	     954	   1137408 ns/op	1843.80 MB/s	    5459 B/op	     173 allocs/op
+// BenchmarkGoFuseRead-2          	    1327	    798635 ns/op	2625.92 MB/s	    5072 B/op	     169 allocs/op
+// BenchmarkGoFuseStat            	    1530	    750944 ns/op
+// BenchmarkGoFuseStat-2          	    8455	    120091 ns/op
+// BenchmarkGoFuseReaddir         	     741	   1561004 ns/op
+// BenchmarkGoFuseReaddir-2       	    2508	    599821 ns/op
+//
+// singleReader = false
+//
+// BenchmarkGoFuseRead            	    1890	    671576 ns/op	3122.73 MB/s	    5393 B/op	     136 allocs/op
+// BenchmarkGoFuseRead-2          	    2948	    429578 ns/op	4881.88 MB/s	   32235 B/op	     157 allocs/op
+// BenchmarkGoFuseStat            	    7886	    153546 ns/op
+// BenchmarkGoFuseStat-2          	    9310	    121332 ns/op
+// BenchmarkGoFuseReaddir         	    4074	    361568 ns/op
+// BenchmarkGoFuseReaddir-2       	    3511	    319765 ns/op
 func (ms *Server) loop(exitIdle bool) {
 	defer ms.loops.Done()
 exit:
@@ -529,15 +562,22 @@ func (ms *Server) handleRequest(req *request) Status {
 
 	errNo := ms.write(req)
 	if errNo != 0 {
-		// Unless debugging is enabled, ignore ENOENT for INTERRUPT responses
-		// which indicates that the referred request is no longer known by the
-		// kernel. This is a normal if the referred request already has
-		// completed.
-		if ms.opts.Debug || !(req.inHeader.Opcode == _OP_INTERRUPT && errNo == ENOENT) {
+		// Ignore ENOENT for INTERRUPT responses which
+		// indicates that the referred request is no longer
+		// known by the kernel. This is a normal if the
+		// referred request already has completed.
+		//
+		// Ignore ENOENT for RELEASE responses.  When the FS
+		// is unmounted directly after a file close, the
+		// device can go away while we are still processing
+		// RELEASE. This is because RELEASE is analogous to
+		// FORGET, and is not synchronized with the calling
+		// process, but does require a response.
+		if ms.opts.Debug || !(errNo == ENOENT && (req.inHeader.Opcode == _OP_INTERRUPT ||
+			req.inHeader.Opcode == _OP_RELEASE)) {
 			ms.opts.Logger.Printf("writer: Write/Writev failed, err: %v. opcode: %v",
 				errNo, operationName(req.inHeader.Opcode))
 		}
-
 	}
 	ms.returnRequest(req)
 	return Status(errNo)

fuse/server_linux.go

@@ -8,6 +8,8 @@ import (
 	"syscall"
 )
 
+const useSingleReader = false
+
 func (ms *Server) systemWrite(req *request, header []byte) Status {
 	if req.flatDataSize() == 0 {
 		err := handleEINTR(func() error {

fuse/server_darwin.go → fuse/server_unix.go

-// Copyright 2016 the Go-FUSE Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
+//go:build !linux
 
 package fuse
 
 import (
-	"syscall"
+	"golang.org/x/sys/unix"
 )
 
+// OSX and FreeBSD has races when multiple routines read
+// from the FUSE device: on unmount, sometime some reads
+// do not error-out, meaning that unmount will hang.
+const useSingleReader = true
+
 func (ms *Server) systemWrite(req *request, header []byte) Status {
 	if req.flatDataSize() == 0 {
 		err := handleEINTR(func() error {
-			_, err := syscall.Write(ms.mountFd, header)
+			_, err := unix.Write(ms.mountFd, header)
 			return err
 		})
 		return ToStatus(err)

fuse/splice_freebsd.go

+package fuse
+
+import "fmt"
+
+func (s *Server) setSplice() {
+	s.canSplice = false
+}
+
+func (ms *Server) trySplice(header []byte, req *request, fdData *readResultFd) error {
+	return fmt.Errorf("unimplemented")
+}

fuse/syscall_darwin.go

@@ -6,49 +6,10 @@ package fuse
 
 import (
 	"bytes"
-	"os"
 	"syscall"
 	"unsafe"
 )
 
-// TODO - move these into Go's syscall package.
-
-func sys_writev(fd int, iovecs *syscall.Iovec, cnt int) (n int, err error) {
-	n1, _, e1 := syscall.Syscall(
-		syscall.SYS_WRITEV,
-		uintptr(fd), uintptr(unsafe.Pointer(iovecs)), uintptr(cnt))
-	n = int(n1)
-	if e1 != 0 {
-		err = syscall.Errno(e1)
-	}
-	return
-}
-
-func writev(fd int, packet [][]byte) (n int, err error) {
-	iovecs := make([]syscall.Iovec, 0, len(packet))
-
-	for _, v := range packet {
-		if len(v) == 0 {
-			continue
-		}
-		vec := syscall.Iovec{
-			Base: &v[0],
-		}
-		vec.SetLen(len(v))
-		iovecs = append(iovecs, vec)
-	}
-
-	sysErr := handleEINTR(func() error {
-		var err error
-		n, err = sys_writev(fd, &iovecs[0], len(iovecs))
-		return err
-	})
-	if sysErr != nil {
-		err = os.NewSyscallError("writev", sysErr)
-	}
-	return n, err
-}
-
 func getxattr(path string, attr string, dest []byte) (sz int, errno int) {
 	pathBs := syscall.StringBytePtr(path)
 	attrBs := syscall.StringBytePtr(attr)

fuse/syscall_linux.go

@@ -5,45 +5,10 @@
 package fuse
 
 import (
-	"os"
-	"syscall"
-	"unsafe"
+	"golang.org/x/sys/unix"
 )
 
-// TODO - move these into Go's syscall package.
-
-func sys_writev(fd int, iovecs *syscall.Iovec, cnt int) (n int, err error) {
-	n1, _, e1 := syscall.Syscall(
-		syscall.SYS_WRITEV,
-		uintptr(fd), uintptr(unsafe.Pointer(iovecs)), uintptr(cnt))
-	n = int(n1)
-	if e1 != 0 {
-		err = syscall.Errno(e1)
-	}
-	return n, err
-}
-
 func writev(fd int, packet [][]byte) (n int, err error) {
-	iovecs := make([]syscall.Iovec, 0, len(packet))
-
-	for _, v := range packet {
-		if len(v) == 0 {
-			continue
-		}
-		vec := syscall.Iovec{
-			Base: &v[0],
-		}
-		vec.SetLen(len(v))
-		iovecs = append(iovecs, vec)
-	}
-
-	sysErr := handleEINTR(func() error {
-		var err error
-		n, err = sys_writev(fd, &iovecs[0], len(iovecs))
-		return err
-	})
-	if sysErr != nil {
-		err = os.NewSyscallError("writev", sysErr)
-	}
-	return n, err
+	n, err = unix.Writev(fd, packet)
+	return
 }

fuse/syscall_unix.go

+//go:build !linux
+
+package fuse
+
+import (
+	"os"
+	"syscall"
+	"unsafe"
+)
+
+func sys_writev(fd int, iovecs *syscall.Iovec, cnt int) (n int, err error) {
+	n1, _, e1 := syscall.Syscall(
+		syscall.SYS_WRITEV,
+		uintptr(fd), uintptr(unsafe.Pointer(iovecs)), uintptr(cnt))
+	n = int(n1)
+	if e1 != 0 {
+		err = syscall.Errno(e1)
+	}
+	return n, err
+}
+
+// Until golang.orgx/sys/unix provides Writev for Darwin and FreeBSD,
+// keep the syscall wrapping funcitons.
+func writev(fd int, packet [][]byte) (n int, err error) {
+	iovecs := make([]syscall.Iovec, 0, len(packet))
+
+	for _, v := range packet {
+		if len(v) == 0 {
+			continue
+		}
+		vec := syscall.Iovec{
+			Base: &v[0],
+		}
+		vec.SetLen(len(v))
+		iovecs = append(iovecs, vec)
+	}
+
+	sysErr := handleEINTR(func() error {
+		var err error
+		n, err = sys_writev(fd, &iovecs[0], len(iovecs))
+		return err
+	})
+	if sysErr != nil {
+		err = os.NewSyscallError("writev", sysErr)
+	}
+	return n, err
+}

fuse/types.go

@@ -44,9 +44,6 @@ const (
 	// ENOSYS Function not implemented
 	ENOSYS = Status(syscall.ENOSYS)
 
-	// ENODATA No data available
-	ENODATA = Status(syscall.ENODATA)
-
 	// ENOTDIR Not a directory
 	ENOTDIR = Status(syscall.ENOTDIR)
 
@@ -259,12 +256,13 @@ const (
 	FOPEN_STREAM                 = (1 << 4)
 	FOPEN_NOFLUSH                = (1 << 5)
 	FOPEN_PARALLEL_DIRECT_WRITES = (1 << 6)
+	FOPEN_PASSTHROUGH            = (1 << 7)
 )
 
 type OpenOut struct {
 	Fh        uint64
 	OpenFlags uint32
-	Padding   uint32
+	BackingID int32
 }
 
 // To be set in InitIn/InitOut.Flags.
@@ -301,12 +299,16 @@ const (
 	CAP_MAX_PAGES        = (1 << 22)
 	CAP_CACHE_SYMLINKS   = (1 << 23)
 
+	/* bits 24..31 differ across linux and mac */
 	/* bits 32..63 get shifted down 32 bits into the Flags2 field */
-	CAP_SECURITY_CTX      = (1 << 32)
-	CAP_HAS_INODE_DAX     = (1 << 33)
-	CAP_CREATE_SUPP_GROUP = (1 << 34)
-	CAP_HAS_EXPIRE_ONLY   = (1 << 35)
-	CAP_DIRECT_IO_RELAX   = (1 << 36)
+	CAP_SECURITY_CTX         = (1 << 32)
+	CAP_HAS_INODE_DAX        = (1 << 33)
+	CAP_CREATE_SUPP_GROUP    = (1 << 34)
+	CAP_HAS_EXPIRE_ONLY      = (1 << 35)
+	CAP_DIRECT_IO_ALLOW_MMAP = (1 << 36)
+	CAP_PASSTHROUGH          = (1 << 37)
+	CAP_NO_EXPORT_SUPPORT    = (1 << 38)
+	CAP_HAS_RESEND           = (1 << 39)
 )
 
 type InitIn struct {
@@ -320,6 +322,10 @@ type InitIn struct {
 	Unused       [11]uint32
 }
 
+func (i *InitIn) Flags64() uint64 {
+	return uint64(i.Flags) | uint64(i.Flags2)<<32
+}
+
 type InitOut struct {
 	Major               uint32
 	Minor               uint32
@@ -332,7 +338,12 @@ type InitOut struct {
 	MaxPages            uint16
 	Padding             uint16
 	Flags2              uint32
-	Unused              [7]uint32
+	MaxStackDepth       uint32
+	Unused              [6]uint32
+}
+
+func (o *InitOut) Flags64() uint64 {
+	return uint64(o.Flags) | uint64(o.Flags2)<<32
 }
 
 type _CuseInitIn struct {
@@ -519,6 +530,7 @@ const (
 	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
+	NOTIFY_RESEND         = -7
 
 // NOTIFY_CODE_MAX     = -6
 )
@@ -693,3 +705,79 @@ func (lk *FileLock) FromFlockT(flockT *syscall.Flock_t) {
 	}
 	lk.Pid = uint32(flockT.Pid)
 }
+
+const (
+	// Mask for GetAttrIn.Flags. If set, GetAttrIn has a file handle set.
+	FUSE_GETATTR_FH = (1 << 0)
+)
+
+type GetAttrIn struct {
+	InHeader
+
+	Flags_ uint32
+	Dummy  uint32
+	Fh_    uint64
+}
+
+// Flags accesses the flags. This is a method, because OSXFuse does not
+// have GetAttrIn flags.
+func (g *GetAttrIn) Flags() uint32 {
+	return g.Flags_
+}
+
+// Fh accesses the file handle. This is a method, because OSXFuse does not
+// have GetAttrIn flags.
+func (g *GetAttrIn) Fh() uint64 {
+	return g.Fh_
+}
+
+type MknodIn struct {
+	InHeader
+
+	// Mode to use, including the Umask value
+	Mode    uint32
+	Rdev    uint32
+	Umask   uint32
+	Padding uint32
+}
+
+type CreateIn struct {
+	InHeader
+	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 ReadIn struct {
+	InHeader
+	Fh        uint64
+	Offset    uint64
+	Size      uint32
+	ReadFlags uint32
+	LockOwner uint64
+	Flags     uint32
+	Padding   uint32
+}
+
+type WriteIn struct {
+	InHeader
+	Fh         uint64
+	Offset     uint64
+	Size       uint32
+	WriteFlags uint32
+	LockOwner  uint64
+	Flags      uint32
+	Padding    uint32
+}
+
+// Data for registering a file as backing an inode.
+type BackingMap struct {
+	Fd      int32
+	Flags   uint32
+	padding uint64
+}

fuse/types_darwin.go

@@ -9,6 +9,7 @@ import (
 )
 
 const (
+	ENODATA = Status(syscall.ENODATA)
 	ENOATTR = Status(syscall.ENOATTR) // ENOATTR is not defined for all GOOS.
 
 	// EREMOTEIO is not supported on Darwin.
@@ -61,73 +62,6 @@ const (
 	FOPEN_PURGE_UBC  = (1 << 31)
 )
 
-// compat with linux.
-const (
-	// Mask for GetAttrIn.Flags. If set, GetAttrIn has a file handle set.
-	FUSE_GETATTR_FH = (1 << 0)
-)
-
-type GetAttrIn struct {
-	InHeader
-
-	Flags_ uint32
-	Dummy  uint32
-	Fh_    uint64
-}
-
-func (g *GetAttrIn) Flags() uint32 {
-	return g.Flags_
-}
-
-func (g *GetAttrIn) Fh() uint64 {
-	return g.Fh_
-}
-
-// Uses OpenIn struct for create.
-type CreateIn struct {
-	InHeader
-	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 MknodIn struct {
-	InHeader
-
-	// Mode to use, including the Umask value
-	Mode    uint32
-	Rdev    uint32
-	Umask   uint32
-	Padding uint32
-}
-
-type ReadIn struct {
-	InHeader
-	Fh        uint64
-	Offset    uint64
-	Size      uint32
-	ReadFlags uint32
-	LockOwner uint64
-	Flags     uint32
-	Padding   uint32
-}
-
-type WriteIn struct {
-	InHeader
-	Fh         uint64
-	Offset     uint64
-	Size       uint32
-	WriteFlags uint32
-	LockOwner  uint64
-	Flags      uint32
-	Padding    uint32
-}
-
 type SetXAttrIn struct {
 	InHeader
 	Size     uint32
@@ -191,3 +125,7 @@ func (s *StatfsOut) FromStatfsT(statfs *syscall.Statfs_t) {
 		s.Bavail /= adj
 	}
 }
+
+func (o *InitOut) setFlags(flags uint64) {
+	o.Flags = uint32(flags)
+}

fuse/types_freebsd.go

+package fuse
+
+import "syscall"
+
+const (
+	ENOATTR = Status(syscall.ENOATTR)
+	ENODATA = Status(syscall.EIO)
+)
+
+const (
+	CAP_NO_OPENDIR_SUPPORT = (1 << 24)
+
+	// The higher capabilities are not currently defined by FreeBSD.
+	// Ref: https://cgit.freebsd.org/src/tree/sys/fs/fuse/fuse_kernel.h
+
+	// CAP_RENAME_SWAP only exists on OSX.
+	CAP_RENAME_SWAP = 0x0
+
+	// CAP_EXPLICIT_INVAL_DATA is not supported on FreeBSD.
+	CAP_EXPLICIT_INVAL_DATA = 0x0
+)
+
+func (s *StatfsOut) FromStatfsT(statfs *syscall.Statfs_t) {
+	s.Blocks = statfs.Blocks
+	s.Bsize = uint32(statfs.Bsize)
+	s.Bfree = statfs.Bfree
+	s.Bavail = uint64(statfs.Bavail)
+	s.Files = statfs.Files
+	s.Ffree = uint64(statfs.Ffree)
+	s.Frsize = uint32(statfs.Bsize)
+	s.NameLen = uint32(statfs.Namemax)
+}

fuse/types_linux.go

@@ -9,6 +9,7 @@ import (
 )
 
 const (
+	ENODATA = Status(syscall.ENODATA)
 	ENOATTR = Status(syscall.ENODATA) // On Linux, ENOATTR is an alias for ENODATA.
 
 	// EREMOTEIO Remote I/O error
@@ -22,124 +23,18 @@ const (
 const (
 	CAP_NO_OPENDIR_SUPPORT  = (1 << 24)
 	CAP_EXPLICIT_INVAL_DATA = (1 << 25)
-	CAP_MAP_ALIGNMENT       = (1 << 26)
-	CAP_SUBMOUNTS           = (1 << 27)
-	CAP_HANDLE_KILLPRIV_V2  = (1 << 28)
-	CAP_SETXATTR_EXT        = (1 << 29)
-	CAP_INIT_EXT            = (1 << 30)
-	CAP_INIT_RESERVED       = (1 << 31)
+
+	CAP_MAP_ALIGNMENT      = (1 << 26)
+	CAP_SUBMOUNTS          = (1 << 27)
+	CAP_HANDLE_KILLPRIV_V2 = (1 << 28)
+	CAP_SETXATTR_EXT       = (1 << 29)
+	CAP_INIT_EXT           = (1 << 30)
+	CAP_INIT_RESERVED      = (1 << 31)
 
 	// CAP_RENAME_SWAP only exists on OSX.
 	CAP_RENAME_SWAP = 0x0
 )
 
-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
-	Owner
-	Rdev uint32
-
-	// Blksize is the preferred size for file system operations.
-	Blksize uint32
-	Padding uint32
-}
-
-type SetAttrIn struct {
-	SetAttrInCommon
-}
-
-const (
-	// Mask for GetAttrIn.Flags. If set, GetAttrIn has a file handle set.
-	FUSE_GETATTR_FH = (1 << 0)
-)
-
-type GetAttrIn struct {
-	InHeader
-
-	Flags_ uint32
-	Dummy  uint32
-	Fh_    uint64
-}
-
-// Flags accesses the flags. This is a method, because OSXFuse does not
-// have GetAttrIn flags.
-func (g *GetAttrIn) Flags() uint32 {
-	return g.Flags_
-}
-
-// Fh accesses the file handle. This is a method, because OSXFuse does not
-// have GetAttrIn flags.
-func (g *GetAttrIn) Fh() uint64 {
-	return g.Fh_
-}
-
-type CreateIn struct {
-	InHeader
-	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 MknodIn struct {
-	InHeader
-
-	// Mode to use, including the Umask value
-	Mode    uint32
-	Rdev    uint32
-	Umask   uint32
-	Padding uint32
-}
-
-type ReadIn struct {
-	InHeader
-	Fh        uint64
-	Offset    uint64
-	Size      uint32
-	ReadFlags uint32
-	LockOwner uint64
-	Flags     uint32
-	Padding   uint32
-}
-
-type WriteIn struct {
-	InHeader
-	Fh         uint64
-	Offset     uint64
-	Size       uint32
-	WriteFlags uint32
-	LockOwner  uint64
-	Flags      uint32
-	Padding    uint32
-}
-
-type SetXAttrIn struct {
-	InHeader
-	Size  uint32
-	Flags uint32
-}
-
-type GetXAttrIn struct {
-	InHeader
-	Size    uint32
-	Padding uint32
-}
-
 func (s *StatfsOut) FromStatfsT(statfs *syscall.Statfs_t) {
 	s.Blocks = statfs.Blocks
 	s.Bsize = uint32(statfs.Bsize)
@@ -150,3 +45,8 @@ func (s *StatfsOut) FromStatfsT(statfs *syscall.Statfs_t) {
 	s.Frsize = uint32(statfs.Frsize)
 	s.NameLen = uint32(statfs.Namelen)
 }
+
+func (o *InitOut) setFlags(flags uint64) {
+	o.Flags = uint32(flags) | CAP_INIT_EXT
+	o.Flags2 = uint32(flags >> 32)
+}

fuse/types_unix.go

+//go:build !darwin
+
+package fuse
+
+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
+	Owner
+	Rdev uint32
+
+	// Blksize is the preferred size for file system operations.
+	Blksize uint32
+	Padding uint32
+}
+
+type SetAttrIn struct {
+	SetAttrInCommon
+}
+
+type SetXAttrIn struct {
+	InHeader
+	Size  uint32
+	Flags uint32
+}
+
+type GetXAttrIn struct {
+	InHeader
+	Size    uint32
+	Padding uint32
+}

go.mod

@@ -7,4 +7,4 @@ require (
 	golang.org/x/sys v0.0.0-20220520151302-bc2c85ada10a
 )
 
-go 1.13
+go 1.17

internal/fallocate/fallocate.go

+package fallocate
+
+// Fallocate is a wrapper around fallocate syscall.
+// On Linux, it is a wrapper around fallocate(2).
+// On Darwin, it is a wrapper around fnctl(2).
+// On FreeBSD, it is a wrapper around posix_fallocate(2).
+func Fallocate(fd int, mode uint32, off int64, len int64) (err error) {
+	return fallocate(fd, mode, off, len)
+}

internal/fallocate/fallocate_darwin.go

+package fallocate
+
+import (
+	"syscall"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+func fallocate(fd int, mode uint32, off int64, len int64) error {
+	// TODO: Handle `mode` parameter.
+	_ = mode
+
+	// From `man fcntl` on OSX:
+	//     The F_PREALLOCATE command operates on the following structure:
+	//
+	//             typedef struct fstore {
+	//                 u_int32_t fst_flags;      /* IN: flags word */
+	//                 int       fst_posmode;    /* IN: indicates offset field */
+	//                 off_t     fst_offset;     /* IN: start of the region */
+	//                 off_t     fst_length;     /* IN: size of the region */
+	//                 off_t     fst_bytesalloc; /* OUT: number of bytes allocated */
+	//             } fstore_t;
+	//
+	//     The flags (fst_flags) for the F_PREALLOCATE command are as follows:
+	//
+	//           F_ALLOCATECONTIG   Allocate contiguous space.
+	//
+	//           F_ALLOCATEALL      Allocate all requested space or no space at all.
+	//
+	//     The position modes (fst_posmode) for the F_PREALLOCATE command indicate how to use the offset field.  The modes are as fol-
+	//     lows:
+	//
+	//           F_PEOFPOSMODE   Allocate from the physical end of file.
+	//
+	//           F_VOLPOSMODE    Allocate from the volume offset.
+
+	k := struct {
+		Flags      uint32 // u_int32_t
+		Posmode    int64  // int
+		Offset     int64  // off_t
+		Length     int64  // off_t
+		Bytesalloc int64  // off_t
+	}{
+		0,
+		0,
+		int64(off),
+		int64(len),
+		0,
+	}
+	_, _, errno := unix.Syscall(syscall.SYS_FCNTL, uintptr(fd), uintptr(unix.F_PREALLOCATE), uintptr(unsafe.Pointer(&k)))
+	if errno != 0 {
+		return errno
+	}
+	return nil
+}

internal/fallocate/fallocate_freebsd.go

+package fallocate
+
+import (
+	"golang.org/x/sys/unix"
+)
+
+func fallocate(fd int, mode uint32, off int64, len int64) error {
+	// Ignore mode
+	_ = mode
+	ret, _, _ := unix.Syscall(unix.SYS_POSIX_FALLOCATE, uintptr(fd), uintptr(off), uintptr(len))
+	if ret != 0 {
+		return unix.Errno(ret)
+	}
+	return nil
+}

internal/fallocate/fallocate_linux.go

+package fallocate
+
+import "golang.org/x/sys/unix"
+
+func fallocate(fd int, mode uint32, off int64, len int64) error {
+	return unix.Fallocate(fd, mode, off, len)
+}

internal/renameat/renameat.go

@@ -3,6 +3,7 @@ package renameat
 // Renameat is a wrapper around renameat syscall.
 // On Linux, it is a wrapper around renameat2(2).
 // On Darwin, it is a wrapper around renameatx_np(2).
+// On FreeBSD, it is a wrapper around renameat(2).
 func Renameat(olddirfd int, oldpath string, newdirfd int, newpath string, flags uint) (err error) {
 	return renameat(olddirfd, oldpath, newdirfd, newpath, flags)
 }

internal/renameat/renameat_freebsd.go

+package renameat
+
+import "golang.org/x/sys/unix"
+
+const (
+	// Since FreeBSD does not currently privode renameat syscall
+	// beyond POSIX standard like Linux and Darwin do, we borrow
+	// the defination from Linux but reject these non-POSIX flags.
+	RENAME_EXCHANGE = (1 << 1)
+)
+
+func renameat(olddirfd int, oldpath string, newdirfd int, newpath string, flags uint) (err error) {
+	if flags != 0 {
+		return unix.ENOSYS
+	}
+	return unix.Renameat(olddirfd, oldpath, newdirfd, newpath)
+}

posixtest/test.go

@@ -17,6 +17,7 @@ import (
 	"testing"
 
 	"github.com/hanwen/go-fuse/v2/fuse"
+	"github.com/hanwen/go-fuse/v2/internal/fallocate"
 	"golang.org/x/sys/unix"
 )
 
@@ -47,6 +48,32 @@ var All = map[string]func(*testing.T, string){
 	"DirSeek":                    DirSeek,
 	"FcntlFlockSetLk":            FcntlFlockSetLk,
 	"FcntlFlockLocksFile":        FcntlFlockLocksFile,
+	"SetattrSymlink":             SetattrSymlink,
+}
+
+func SetattrSymlink(t *testing.T, mnt string) {
+	l := filepath.Join(mnt, "link")
+	if err := os.Symlink("doesnotexist", l); err != nil {
+		t.Fatalf("symlink: %v", err)
+	}
+
+	tvs := []unix.Timeval{
+		{Sec: 42, Usec: 1},
+		{Sec: 43, Usec: 2},
+	}
+	if err := unix.Lutimes(l, tvs); err != nil {
+		t.Fatalf("Lutimes: %v", err)
+	}
+
+	var st unix.Stat_t
+	if err := unix.Lstat(l, &st); err != nil {
+		t.Fatalf("Lstat: %v", err)
+	}
+
+	if st.Mtim.Sec != 43 {
+		// Can't check atime; it's hard to prevent implicit readlink calls.
+		t.Fatalf("got mtime %v, want 43", st.Mtim)
+	}
 }
 
 func DirectIO(t *testing.T, mnt string) {
@@ -272,7 +299,7 @@ func FstatDeleted(t *testing.T, mnt string) {
 	const iMax = 9
 	type file struct {
 		fd int
-		st syscall.Stat_t
+		st unix.Stat_t
 	}
 	files := make(map[int]file)
 	for i := 0; i <= iMax; i++ {
@@ -283,8 +310,8 @@ func FstatDeleted(t *testing.T, mnt string) {
 		if err != nil {
 			t.Fatalf("WriteFile: %v", err)
 		}
-		var st syscall.Stat_t
-		err = syscall.Stat(path, &st)
+		var st unix.Stat_t
+		err = unix.Stat(path, &st)
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -303,14 +330,14 @@ func FstatDeleted(t *testing.T, mnt string) {
 	}
 	// Fstat in random order
 	for _, v := range files {
-		var st syscall.Stat_t
-		err := syscall.Fstat(v.fd, &st)
+		var st unix.Stat_t
+		err := unix.Fstat(v.fd, &st)
 		if err != nil {
 			t.Fatal(err)
 		}
 		// Ignore ctime, changes on unlink
-		v.st.Ctim = syscall.Timespec{}
-		st.Ctim = syscall.Timespec{}
+		v.st.Ctim = unix.Timespec{}
+		st.Ctim = unix.Timespec{}
 		// Nlink value should have dropped to zero
 		v.st.Nlink = 0
 		// Rest should stay the same
@@ -607,7 +634,7 @@ func OpenAt(t *testing.T, mnt string) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	fd, err := syscall.Openat(dirfd, "file1", syscall.O_CREAT, 0700)
+	fd, err := unix.Openat(dirfd, "file1", syscall.O_CREAT, 0700)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -624,7 +651,7 @@ func Fallocate(t *testing.T, mnt string) {
 		t.Fatalf("OpenFile failed: %v", err)
 	}
 	defer rwFile.Close()
-	err = syscall.Fallocate(int(rwFile.Fd()), 0, 1024, 4096)
+	err = fallocate.Fallocate(int(rwFile.Fd()), 0, 1024, 4096)
 	if err != nil {
 		t.Fatalf("Fallocate failed: %v", err)
 	}
@@ -661,12 +688,11 @@ func FcntlFlockSetLk(t *testing.T, mnt string) {
 		}
 		defer f2.Close()
 		lk := syscall.Flock_t{}
-		if err := syscall.FcntlFlock(f2.Fd(), unix.F_OFD_GETLK, &lk); err != nil {
+		if err := sysFcntlFlockGetOFDLock(f2.Fd(), &lk); err != nil {
 			t.Errorf("FcntlFlock failed: %v", err)
 		}
 		if lk.Type != syscall.F_WRLCK {
 			t.Errorf("got lk.Type=%v, want %v", lk.Type, syscall.F_WRLCK)
-
 		}
 	}
 }

posixtest/test_darwin.go

+package posixtest
+
+import "syscall"
+
+// Exist at least from macOS Siera 10.12
+const sys_F_OFD_GETLK = 92
+
+func sysFcntlFlockGetOFDLock(fd uintptr, lk *syscall.Flock_t) error {
+	return syscall.FcntlFlock(fd, sys_F_OFD_GETLK, lk)
+}

posixtest/test_freebsd.go

+package posixtest
+
+import (
+	"syscall"
+	"unsafe"
+)
+
+// Since FreeBSD doesn't implement the F_OFD_GETLK, to mimic its behaviour,
+// we here first fork the process via syscall, and execute fcntl(2) in the child
+// process to get the lock info. Then we send the lock info via pipe back to
+// the parent test process.
+
+func sysFcntlFlockGetOFDLock(fd uintptr, lk *syscall.Flock_t) error {
+	pipefd := make([]int, 2)
+	err := syscall.Pipe(pipefd)
+	if err != nil {
+		return err
+	}
+	pid, _, err := syscall.Syscall(syscall.SYS_FORK, 0, 0, 0)
+	if pid == 0 { // child process
+		syscall.Close(pipefd[0]) // close read end
+		var clk syscall.Flock_t
+		// Here we must give a vaild lock type, or fcntl(2) will return
+		// EINVAL. And it should be different from what we set earlier
+		// in the test. Here we set to F_RDLOCK.
+		clk.Type = syscall.F_RDLCK
+		syscall.FcntlFlock(fd, syscall.F_GETLK, &clk)
+		syscall.Syscall(syscall.SYS_WRITE, uintptr(pipefd[1]), uintptr(unsafe.Pointer(&clk)), unsafe.Sizeof(clk))
+		syscall.Exit(0)
+	} else if pid > 0 { // parent process
+		syscall.Close(pipefd[1]) // close write end
+		buf := make([]byte, unsafe.Sizeof(*lk))
+		syscall.Read(pipefd[0], buf)
+		*lk = *((*syscall.Flock_t)(unsafe.Pointer(&buf[0])))
+		syscall.Close(pipefd[0]) // close read end
+	} else {
+		return err
+	}
+	return nil
+}

posixtest/test_linux.go

+package posixtest
+
+import (
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+func sysFcntlFlockGetOFDLock(fd uintptr, lk *syscall.Flock_t) error {
+	return syscall.FcntlFlock(fd, unix.F_OFD_GETLK, lk)
+}

posixtest/xfstests.go

@@ -11,6 +11,10 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+type dirent struct {
+	unix.Dirent
+}
+
 // DirSeek tests that seeking on a directory works for
 // https://github.com/hanwen/go-fuse/issues/344 .
 //
@@ -47,7 +51,7 @@ func DirSeek(t *testing.T, mnt string) {
 
 	total := 0
 	for {
-		n, err := unix.Getdents(fd, buf)
+		n, err := unix.ReadDirent(fd, buf)
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -55,24 +59,24 @@ func DirSeek(t *testing.T, mnt string) {
 			break
 		}
 		for bpos := 0; bpos < n; total++ {
-			d := (*unix.Dirent)(unsafe.Pointer(&buf[bpos]))
+			d := (*dirent)(unsafe.Pointer(&buf[bpos]))
 			if total > historyLen {
 				t.Fatal("too many files")
 			}
 			for i := 0; i < total; i++ {
-				if offHistory[i] == d.Off {
+				if offHistory[i] == d.off() {
 					t.Errorf("entries %d and %d gave duplicate d.Off %d",
-						i, total, d.Off)
+						i, total, d.off())
 				}
 			}
-			offHistory[total] = d.Off
-			inoHistory[total] = d.Ino
+			offHistory[total] = d.off()
+			inoHistory[total] = d.ino()
 			bpos += int(d.Reclen)
 		}
 	}
 
 	// check if seek works correctly
-	d := (*unix.Dirent)(unsafe.Pointer(&buf[0]))
+	d := (*dirent)(unsafe.Pointer(&buf[0]))
 	for i := total - 1; i >= 0; i-- {
 		var seekTo int64
 		if i > 0 {
@@ -83,7 +87,7 @@ func DirSeek(t *testing.T, mnt string) {
 			t.Fatal(err)
 		}
 
-		n, err := unix.Getdents(fd, buf)
+		n, err := unix.ReadDirent(fd, buf)
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -92,9 +96,9 @@ func DirSeek(t *testing.T, mnt string) {
 			continue
 		}
 
-		if d.Ino != inoHistory[i] {
+		if d.ino() != inoHistory[i] {
 			t.Errorf("entry %d has inode %d, expected %d",
-				i, d.Ino, inoHistory[i])
+				i, d.ino(), inoHistory[i])
 		}
 	}
 }

posixtest/xfstests_darwin.go

+package posixtest
+
+func (d *dirent) off() int64 {
+	return int64(d.Seekoff)
+}
+
+func (d *dirent) ino() uint64 {
+	return d.Ino
+}

posixtest/xfstests_freebsd.go

+package posixtest
+
+func (d *dirent) off() int64 {
+	return d.Off
+}
+
+func (d *dirent) ino() uint64 {
+	return d.Fileno
+}

posixtest/xfstests_linux.go

+package posixtest
+
+func (d *dirent) off() int64 {
+	return d.Off
+}
+
+func (d *dirent) ino() uint64 {
+	return d.Ino
+}

splice/copy.go

+//go:build linux
+
 // Copyright 2016 the Go-FUSE Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

splice/copy_test.go

+//go:build linux
+
 // Copyright 2016 the Go-FUSE Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

splice/pair.go

+//go:build linux
+
 // Copyright 2016 the Go-FUSE Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

splice/pair_darwin.go

-// Copyright 2016 the Go-FUSE Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package splice
-
-func (p *Pair) LoadFromAt(fd uintptr, sz int, off int64) (int, error) {
-	panic("not implemented")
-	return 0, nil
-}
-
-func (p *Pair) LoadFrom(fd uintptr, sz int) (int, error) {
-	panic("not implemented")
-	return 0, nil
-}
-
-func (p *Pair) WriteTo(fd uintptr, n int) (int, error) {
-	panic("not implemented")
-	return 0, nil
-}

splice/pool.go

+//go:build linux
+
 // Copyright 2016 the Go-FUSE Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

splice/splice.go

+//go:build linux
+
 // Copyright 2016 the Go-FUSE Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

splice/splice_test.go

+//go:build linux
+
 // Copyright 2016 the Go-FUSE Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.