Expose Gid field in OpContext

This makes the directmount code path pick up this workaround,
which fixes mounting jacobsa/fuse based file systems (which
do not set the FSName explicitly, like the samples) in Docker:

dd426a02d6/mount_config.go (L208-L220)

fixes https://github.com/jacobsa/fuse/issues/166

connection.go

@@ -39,17 +39,19 @@ var contextKey interface{} = contextKeyType(0)
 //
 // As of 2015-03-26, the behavior in the kernel is:
 //
-//   - (http://goo.gl/bQ1f1i, http://goo.gl/HwBrR6) Set the local variable
-//     ra_pages to be init_response->max_readahead divided by the page size.
+//   - (https://tinyurl.com/2eakn5e9, https://tinyurl.com/mry9e33d) Set the
+//     local variable ra_pages to be init_response->max_readahead divided by
+//     the page size.
 //
-//   - (http://goo.gl/gcIsSh, http://goo.gl/LKV2vA) Set
-//     backing_dev_info::ra_pages to the min of that value and what was sent
-//     in the request's max_readahead field.
+//   - (https://tinyurl.com/2eakn5e9, https://tinyurl.com/mbpshk8h) Set
+//     backing_dev_info::ra_pages to the min of that value and what was sent in
+//     the request's max_readahead field.
 //
-//   - (http://goo.gl/u2SqzH) Use backing_dev_info::ra_pages when deciding
-//     how much to read ahead.
+//   - (https://tinyurl.com/57hpfu4x) Use backing_dev_info::ra_pages when
+//     deciding how much to read ahead.
 //
-//   - (http://goo.gl/JnhbdL) Don't read ahead at all if that field is zero.
+//   - (https://tinyurl.com/ywhfcfte) Don't read ahead at all if that field is
+//     zero.
 //
 // Reading a page at a time is a drag. Ask for a larger size.
 const maxReadahead = 1 << 20
@@ -151,6 +153,7 @@ func (c *Connection) Init() error {
 	cacheSymlinks := initOp.Flags&fusekernel.InitCacheSymlinks > 0
 	noOpenSupport := initOp.Flags&fusekernel.InitNoOpenSupport > 0
 	noOpendirSupport := initOp.Flags&fusekernel.InitNoOpendirSupport > 0
+	readdirplusSupport := initOp.Flags&fusekernel.InitDoReaddirplus > 0
 
 	// Respond to the init op.
 	initOp.Library = c.protocol
@@ -193,8 +196,17 @@ func (c *Connection) Init() error {
 		initOp.Flags |= fusekernel.InitNoOpendirSupport
 	}
 
-	c.Reply(ctx, nil)
-	return nil
+	// Tell the Kernel to allow sending parallel lookup and readdir operations.
+	if c.cfg.EnableParallelDirOps {
+		initOp.Flags |= fusekernel.InitParallelDirOps
+	}
+
+	// Tell the kernel to do readdirplus (readdir+lookup in one call)
+	if c.cfg.UseReadDirPlus && readdirplusSupport {
+		initOp.Flags |= fusekernel.InitDoReaddirplus
+	}
+
+	return c.Reply(ctx, nil)
 }
 
 // Log information for an operation with the given ID. calldepth is the depth
@@ -309,13 +321,13 @@ func (c *Connection) handleInterrupt(fuseID uint64) {
 	defer c.mu.Unlock()
 
 	// NOTE(jacobsa): fuse.txt in the Linux kernel documentation
-	// (https://goo.gl/H55Dnr) defines the kernel <-> userspace protocol for
-	// interrupts.
+	// (https://tinyurl.com/2r4ajuwd) defines the kernel <-> userspace protocol
+	// for interrupts.
 	//
 	// In particular, my reading of it is that an interrupt request cannot be
 	// delivered to userspace before the original request. The part about the
 	// race and EAGAIN appears to be aimed at userspace programs that
-	// concurrently process requests (cf. http://goo.gl/BES2rs).
+	// concurrently process requests (https://tinyurl.com/3euehwfb).
 	//
 	// So in this method if we can't find the ID to be interrupted, it means that
 	// the request has already been replied to.
@@ -369,18 +381,33 @@ func (c *Connection) readMessage() (*buffer.InMessage, error) {
 }
 
 // Write the supplied message to the kernel.
-func (c *Connection) writeMessage(msg []byte) error {
-	// Avoid the retry loop in os.File.Write.
-	n, err := syscall.Write(int(c.dev.Fd()), msg)
+func (c *Connection) writeMessage(outMsg *buffer.OutMessage) error {
+	var err error
+	var n int
+	expectedLen := outMsg.Len()
+	if outMsg.Sglist != nil {
+		if fusekernel.IsPlatformFuseT {
+			// writev is not atomic on macos, restrict to fuse-t platform
+			writeLock.Lock()
+			defer writeLock.Unlock()
+		}
+		n, err = writev(int(c.dev.Fd()), outMsg.Sglist)
+	} else {
+		// Avoid the retry loop in os.File.Write.
+		n, err = syscall.Write(int(c.dev.Fd()), outMsg.OutHeaderBytes())
+	}
+	if err == nil && n != expectedLen {
+		err = fmt.Errorf("Wrote %d bytes; expected %d", n, expectedLen)
+	}
 	if err != nil {
-		return err
+		writeErrMsg := fmt.Sprintf("writeMessage: %v %v", err, outMsg.OutHeaderBytes())
+		if c.errorLogger != nil {
+			c.errorLogger.Print(writeErrMsg)
+		}
+		return fmt.Errorf(writeErrMsg)
 	}
-
-	if n != len(msg) {
-		return fmt.Errorf("Wrote %d bytes; expected %d", n, len(msg))
-	}
-
-	return nil
+	outMsg.Sglist = nil
+	return err
 }
 
 // ReadOp consumes the next op from the kernel process, returning the op and a
@@ -454,7 +481,7 @@ func (c *Connection) shouldLogError(
 			return false
 		}
 	case *fuseops.GetXattrOp, *fuseops.ListXattrOp:
-		if err == syscall.ENODATA || err == syscall.ERANGE {
+		if err == syscall.ENOSYS || err == syscall.ENODATA || err == syscall.ERANGE {
 			return false
 		}
 	case *unknownOp:
@@ -467,11 +494,13 @@ func (c *Connection) shouldLogError(
 	return true
 }
 
+var writeLock sync.Mutex
+
 // Reply replies to an op previously read using ReadOp, with the supplied error
 // (or nil if successful). The context must be the context returned by ReadOp.
 //
 // LOCKS_EXCLUDED(c.mu)
-func (c *Connection) Reply(ctx context.Context, opErr error) {
+func (c *Connection) Reply(ctx context.Context, opErr error) error {
 	// Extract the state we stuffed in earlier.
 	var key interface{} = contextKey
 	foo := ctx.Value(key)
@@ -485,9 +514,25 @@ func (c *Connection) Reply(ctx context.Context, opErr error) {
 	outMsg := state.outMsg
 	fuseID := inMsg.Header().Unique
 
-	// Make sure we destroy the messages when we're done.
-	defer c.putInMessage(inMsg)
-	defer c.putOutMessage(outMsg)
+	defer func() {
+		// Invoke any callbacks set by the FUSE server after the response to the kernel is
+		// complete and before the inMessage and outMessage memory buffers have been freed.
+		callback := c.callbackForOp(op)
+		if callback != nil {
+			callback()
+		}
+
+		// Make sure we destroy the messages when we're done.
+		suppressReuse := false
+		if wr, ok := op.(*fuseops.WriteFileOp); ok {
+			suppressReuse = wr.SuppressReuse
+		}
+
+		if !suppressReuse {
+			c.putInMessage(inMsg)
+		}
+		c.putOutMessage(outMsg)
+	}()
 
 	// Clean up state for this op.
 	c.finishOp(inMsg.Header().Opcode, inMsg.Header().Unique)
@@ -495,7 +540,7 @@ func (c *Connection) Reply(ctx context.Context, opErr error) {
 	// Debug logging
 	if c.debugLogger != nil {
 		if opErr == nil {
-			c.debugLog(fuseID, 1, "-> OK (%s)", describeResponse(op))
+			c.debugLog(fuseID, 1, "-> %s", describeResponse(op))
 		} else {
 			c.debugLog(fuseID, 1, "-> Error: %q", opErr.Error())
 		}
@@ -510,17 +555,31 @@ func (c *Connection) Reply(ctx context.Context, opErr error) {
 	noResponse := c.kernelResponse(outMsg, inMsg.Header().Unique, op, opErr)
 
 	if !noResponse {
-		var err error
-		if outMsg.Sglist != nil {
-			_, err = writev(int(c.dev.Fd()), outMsg.Sglist)
-		} else {
-			err = c.writeMessage(outMsg.OutHeaderBytes())
-		}
-		if err != nil && c.errorLogger != nil {
-			c.errorLogger.Printf("writeMessage: %v %v", err, outMsg.OutHeaderBytes())
-		}
-		outMsg.Sglist = nil
+		c.writeMessage(outMsg)
 	}
+
+	return nil
+}
+
+func (c *Connection) callbackForOp(op interface{}) func() {
+	switch o := op.(type) {
+	case *fuseops.ReadFileOp:
+		return o.Callback
+	case *fuseops.WriteFileOp:
+		return o.Callback
+	}
+	return nil
+}
+
+// Send a notification to the kernel
+// notification must be a pointer to one of fuseops.NotifyXXX structures
+// To avoid a deadlock notifications must not be called in the execution path of a related filesytem operation or within any code that could hold a lock that could be needed to execute such an operation. As of kernel 4.18, a "related operation" is a lookup(), symlink(), mknod(), mkdir(), unlink(), rename(), link() or create() request for the parent, and a setattr(), unlink(), rmdir(), rename(), setxattr(), removexattr(), readdir() or readdirplus() request for the inode itself.
+func (c *Connection) Notify(notification interface{}) error {
+	outMsg := c.getOutMessage()
+	defer c.putOutMessage(outMsg)
+	c.kernelNotification(outMsg, notification)
+	outMsg.OutHeader().Len = uint32(outMsg.Len())
+	return c.writeMessage(outMsg)
 }
 
 // Close the connection. Must not be called until operations that were read

conversions.go

@@ -51,15 +51,25 @@ func convertInMessage(
 		}
 
 		o = &fuseops.LookUpInodeOp{
-			Parent:    fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(buf[:n-1]),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Parent: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:   string(buf[:n-1]),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpGetattr:
 		o = &fuseops.GetInodeAttributesOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpSetattr:
@@ -70,8 +80,13 @@ func convertInMessage(
 		}
 
 		to := &fuseops.SetInodeAttributesOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 		o = to
 
@@ -89,7 +104,7 @@ func convertInMessage(
 		}
 
 		if valid&fusekernel.SetattrMode != 0 {
-			mode := convertFileMode(in.Mode)
+			mode := fuseops.ConvertFileMode(in.Mode)
 			to.Mode = &mode
 		}
 
@@ -116,9 +131,14 @@ func convertInMessage(
 		}
 
 		o = &fuseops.ForgetInodeOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			N:         in.Nlookup,
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			N:     in.Nlookup,
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpBatchForget:
@@ -143,8 +163,13 @@ func convertInMessage(
 		}
 
 		o = &fuseops.BatchForgetOp{
-			Entries:   entries,
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Entries: entries,
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpMkdir:
@@ -165,13 +190,18 @@ func convertInMessage(
 			Name:   string(name),
 
 			// On Linux, vfs_mkdir calls through to the inode with at most
-			// permissions and sticky bits set (cf. https://goo.gl/WxgQXk), and fuse
-			// passes that on directly (cf. https://goo.gl/f31aMo). In other words,
-			// the fact that this is a directory is implicit in the fact that the
-			// opcode is mkdir. But we want the correct mode to go through, so ensure
-			// that os.ModeDir is set.
-			Mode:      convertFileMode(in.Mode) | os.ModeDir,
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			// permissions and sticky bits set (https://tinyurl.com/3djx8498), and
+			// fuse passes that on directly (https://tinyurl.com/exezw647). In other
+			// words, the fact that this is a directory is implicit in the fact that
+			// the opcode is mkdir. But we want the correct mode to go through, so
+			// ensure that os.ModeDir is set.
+			Mode: fuseops.ConvertFileMode(in.Mode) | os.ModeDir,
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpMknod:
@@ -188,10 +218,16 @@ func convertInMessage(
 		name = name[:i]
 
 		o = &fuseops.MkNodeOp{
-			Parent:    fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(name),
-			Mode:      convertFileMode(in.Mode),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Parent: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:   string(name),
+			Mode:   fuseops.ConvertFileMode(in.Mode),
+			Rdev:   in.Rdev,
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpCreate:
@@ -208,10 +244,15 @@ func convertInMessage(
 		name = name[:i]
 
 		o = &fuseops.CreateFileOp{
-			Parent:    fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(name),
-			Mode:      convertFileMode(in.Mode),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Parent: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:   string(name),
+			Mode:   fuseops.ConvertFileMode(in.Mode),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpSymlink:
@@ -227,10 +268,15 @@ func convertInMessage(
 		newName, target := names[0:i], names[i+1:len(names)-1]
 
 		o = &fuseops.CreateSymlinkOp{
-			Parent:    fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(newName),
-			Target:    string(target),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Parent: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:   string(newName),
+			Target: string(target),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpRename:
@@ -272,7 +318,12 @@ func convertInMessage(
 			OldName:   string(oldName),
 			NewParent: fuseops.InodeID(in.Newdir),
 			NewName:   string(newName),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpUnlink:
@@ -283,9 +334,14 @@ func convertInMessage(
 		}
 
 		o = &fuseops.UnlinkOp{
-			Parent:    fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(buf[:n-1]),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Parent: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:   string(buf[:n-1]),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpRmdir:
@@ -296,9 +352,14 @@ func convertInMessage(
 		}
 
 		o = &fuseops.RmDirOp{
-			Parent:    fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(buf[:n-1]),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Parent: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:   string(buf[:n-1]),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpOpen:
@@ -311,13 +372,23 @@ func convertInMessage(
 		o = &fuseops.OpenFileOp{
 			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
 			OpenFlags: fusekernel.OpenFlags(in.Flags),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpOpendir:
 		o = &fuseops.OpenDirOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpRead:
@@ -327,11 +398,16 @@ func convertInMessage(
 		}
 
 		to := &fuseops.ReadFileOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Handle:    fuseops.HandleID(in.Fh),
-			Offset:    int64(in.Offset),
-			Size:      int64(in.Size),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode:  fuseops.InodeID(inMsg.Header().Nodeid),
+			Handle: fuseops.HandleID(in.Fh),
+			Offset: int64(in.Offset),
+			Size:   int64(in.Size),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 		if !config.UseVectoredRead {
 			// Use part of the incoming message storage as the read buffer
@@ -340,6 +416,8 @@ func convertInMessage(
 		}
 		o = to
 
+	case fusekernel.OpReaddirplus:
+		fallthrough
 	case fusekernel.OpReaddir:
 		in := (*fusekernel.ReadIn)(inMsg.Consume(fusekernel.ReadInSize(protocol)))
 		if in == nil {
@@ -347,10 +425,16 @@ func convertInMessage(
 		}
 
 		to := &fuseops.ReadDirOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Handle:    fuseops.HandleID(in.Fh),
-			Offset:    fuseops.DirOffset(in.Offset),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode:  fuseops.InodeID(inMsg.Header().Nodeid),
+			Handle: fuseops.HandleID(in.Fh),
+			Offset: fuseops.DirOffset(in.Offset),
+			Plus:   inMsg.Header().Opcode == fusekernel.OpReaddirplus,
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 		o = to
 
@@ -373,8 +457,13 @@ func convertInMessage(
 		}
 
 		o = &fuseops.ReleaseFileHandleOp{
-			Handle:    fuseops.HandleID(in.Fh),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Handle: fuseops.HandleID(in.Fh),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpReleasedir:
@@ -385,8 +474,13 @@ func convertInMessage(
 		}
 
 		o = &fuseops.ReleaseDirHandleOp{
-			Handle:    fuseops.HandleID(in.Fh),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Handle: fuseops.HandleID(in.Fh),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpWrite:
@@ -401,23 +495,45 @@ func convertInMessage(
 		}
 
 		o = &fuseops.WriteFileOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Handle:    fuseops.HandleID(in.Fh),
-			Data:      buf,
-			Offset:    int64(in.Offset),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode:  fuseops.InodeID(inMsg.Header().Nodeid),
+			Handle: fuseops.HandleID(in.Fh),
+			Data:   buf,
+			Offset: int64(in.Offset),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpFsync, fusekernel.OpFsyncdir:
 		type input fusekernel.FsyncIn
 		in := (*input)(inMsg.Consume(unsafe.Sizeof(input{})))
 		if in == nil {
-			return nil, errors.New("Corrupt OpFsync")
+			return nil, errors.New("Corrupt OpFsync/OpFsyncdir")
 		}
 
 		o = &fuseops.SyncFileOp{
+			Inode:  fuseops.InodeID(inMsg.Header().Nodeid),
+			Handle: fuseops.HandleID(in.Fh),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
+		}
+
+	case fusekernel.OpSyncFS:
+		type input fusekernel.SyncFSIn
+		in := (*input)(inMsg.Consume(unsafe.Sizeof(input{})))
+		if in == nil {
+			return nil, errors.New("Corrupt OpSyncFS")
+		}
+
+		o = &fuseops.SyncFSOp{
 			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Handle:    fuseops.HandleID(in.Fh),
 			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
 		}
 
@@ -429,15 +545,25 @@ func convertInMessage(
 		}
 
 		o = &fuseops.FlushFileOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Handle:    fuseops.HandleID(in.Fh),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode:  fuseops.InodeID(inMsg.Header().Nodeid),
+			Handle: fuseops.HandleID(in.Fh),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpReadlink:
 		o = &fuseops.ReadSymlinkOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpStatfs:
@@ -485,10 +611,15 @@ func convertInMessage(
 		}
 
 		o = &fuseops.CreateLinkOp{
-			Parent:    fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(name),
-			Target:    fuseops.InodeID(in.Oldnodeid),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Parent: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:   string(name),
+			Target: fuseops.InodeID(in.Oldnodeid),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpRemovexattr:
@@ -499,9 +630,14 @@ func convertInMessage(
 		}
 
 		o = &fuseops.RemoveXattrOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(buf[:n-1]),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:  string(buf[:n-1]),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 
 	case fusekernel.OpGetxattr:
@@ -519,9 +655,14 @@ func convertInMessage(
 		name = name[:i]
 
 		to := &fuseops.GetXattrOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(name),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:  string(name),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 		o = to
 
@@ -548,8 +689,13 @@ func convertInMessage(
 		}
 
 		to := &fuseops.ListXattrOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 		o = to
 
@@ -584,11 +730,16 @@ func convertInMessage(
 		name, value := payload[:i], payload[i+1:len(payload)]
 
 		o = &fuseops.SetXattrOp{
-			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
-			Name:      string(name),
-			Value:     value,
-			Flags:     in.Flags,
-			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+			Inode: fuseops.InodeID(inMsg.Header().Nodeid),
+			Name:  string(name),
+			Value: value,
+			Flags: in.Flags,
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
 		}
 	case fusekernel.OpFallocate:
 		type input fusekernel.FallocateIn
@@ -598,11 +749,52 @@ func convertInMessage(
 		}
 
 		o = &fuseops.FallocateOp{
+			Inode:  fuseops.InodeID(inMsg.Header().Nodeid),
+			Handle: fuseops.HandleID(in.Fh),
+			Offset: in.Offset,
+			Length: in.Length,
+			Mode:   in.Mode,
+			OpContext: fuseops.OpContext{
+				FuseID: inMsg.Header().Unique,
+				Pid:    inMsg.Header().Pid,
+				Uid:    inMsg.Header().Uid,
+				Gid:    inMsg.Header().Gid,
+			},
+		}
+
+	case fusekernel.OpPoll:
+		type input fusekernel.PollIn
+		in := (*input)(inMsg.Consume(unsafe.Sizeof(input{})))
+		if in == nil {
+			return nil, errors.New("Corrupt OpPoll")
+		}
+
+		o = &fuseops.PollOp{
 			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
 			Handle:    fuseops.HandleID(in.Fh),
+			Kh:        in.Kh,
+			Flags:     fusekernel.PollFlags(in.Flags),
+			Events:    fusekernel.PollEvents(in.Events),
+			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
+		}
+
+	case fusekernel.OpNotifyReply:
+		type input fusekernel.NotifyRetrieveIn
+		in := (*input)(inMsg.Consume(unsafe.Sizeof(input{})))
+		if in == nil {
+			return nil, errors.New("Corrupt OpNotifyReply")
+		}
+
+		buf := inMsg.ConsumeBytes(inMsg.Len())
+		if len(buf) < int(in.Size) {
+			return nil, errors.New("Corrupt OpNotifyReply")
+		}
+
+		o = &fuseops.NotifyRetrieveReplyOp{
+			Inode:     fuseops.InodeID(inMsg.Header().Nodeid),
+			Unique:    inMsg.Header().Unique,
 			Offset:    in.Offset,
-			Length:    in.Length,
-			Mode:      in.Mode,
+			Length:    in.Size,
 			OpContext: fuseops.OpContext{Pid: inMsg.Header().Pid},
 		}
 
@@ -639,6 +831,9 @@ func (c *Connection) kernelResponse(
 	case *fuseops.BatchForgetOp:
 		return true
 
+	case *fuseops.NotifyRetrieveReplyOp:
+		return true
+
 	case *interruptOp:
 		return true
 	}
@@ -682,37 +877,37 @@ func (c *Connection) kernelResponseForOp(
 	case *fuseops.LookUpInodeOp:
 		size := int(fusekernel.EntryOutSize(c.protocol))
 		out := (*fusekernel.EntryOut)(m.Grow(size))
-		convertChildInodeEntry(&o.Entry, out)
+		fuseops.ConvertChildInodeEntry(&o.Entry, out)
 
 	case *fuseops.GetInodeAttributesOp:
 		size := int(fusekernel.AttrOutSize(c.protocol))
 		out := (*fusekernel.AttrOut)(m.Grow(size))
-		out.AttrValid, out.AttrValidNsec = convertExpirationTime(
+		out.AttrValid, out.AttrValidNsec = fuseops.ConvertExpirationTime(
 			o.AttributesExpiration)
-		convertAttributes(o.Inode, &o.Attributes, &out.Attr)
+		fuseops.ConvertAttributes(o.Inode, &o.Attributes, &out.Attr)
 
 	case *fuseops.SetInodeAttributesOp:
 		size := int(fusekernel.AttrOutSize(c.protocol))
 		out := (*fusekernel.AttrOut)(m.Grow(size))
-		out.AttrValid, out.AttrValidNsec = convertExpirationTime(
+		out.AttrValid, out.AttrValidNsec = fuseops.ConvertExpirationTime(
 			o.AttributesExpiration)
-		convertAttributes(o.Inode, &o.Attributes, &out.Attr)
+		fuseops.ConvertAttributes(o.Inode, &o.Attributes, &out.Attr)
 
 	case *fuseops.MkDirOp:
 		size := int(fusekernel.EntryOutSize(c.protocol))
 		out := (*fusekernel.EntryOut)(m.Grow(size))
-		convertChildInodeEntry(&o.Entry, out)
+		fuseops.ConvertChildInodeEntry(&o.Entry, out)
 
 	case *fuseops.MkNodeOp:
 		size := int(fusekernel.EntryOutSize(c.protocol))
 		out := (*fusekernel.EntryOut)(m.Grow(size))
-		convertChildInodeEntry(&o.Entry, out)
+		fuseops.ConvertChildInodeEntry(&o.Entry, out)
 
 	case *fuseops.CreateFileOp:
 		eSize := int(fusekernel.EntryOutSize(c.protocol))
 
 		e := (*fusekernel.EntryOut)(m.Grow(eSize))
-		convertChildInodeEntry(&o.Entry, e)
+		fuseops.ConvertChildInodeEntry(&o.Entry, e)
 
 		oo := (*fusekernel.OpenOut)(m.Grow(int(unsafe.Sizeof(fusekernel.OpenOut{}))))
 		oo.Fh = uint64(o.Handle)
@@ -720,12 +915,12 @@ func (c *Connection) kernelResponseForOp(
 	case *fuseops.CreateSymlinkOp:
 		size := int(fusekernel.EntryOutSize(c.protocol))
 		out := (*fusekernel.EntryOut)(m.Grow(size))
-		convertChildInodeEntry(&o.Entry, out)
+		fuseops.ConvertChildInodeEntry(&o.Entry, out)
 
 	case *fuseops.CreateLinkOp:
 		size := int(fusekernel.EntryOutSize(c.protocol))
 		out := (*fusekernel.EntryOut)(m.Grow(size))
-		convertChildInodeEntry(&o.Entry, out)
+		fuseops.ConvertChildInodeEntry(&o.Entry, out)
 
 	case *fuseops.RenameOp:
 		// Empty response
@@ -740,6 +935,14 @@ func (c *Connection) kernelResponseForOp(
 		out := (*fusekernel.OpenOut)(m.Grow(int(unsafe.Sizeof(fusekernel.OpenOut{}))))
 		out.Fh = uint64(o.Handle)
 
+		if o.CacheDir {
+			out.OpenFlags |= uint32(fusekernel.OpenCacheDir)
+		}
+
+		if o.KeepCache {
+			out.OpenFlags |= uint32(fusekernel.OpenKeepCache)
+		}
+
 	case *fuseops.ReadDirOp:
 		// convertInMessage already set up the destination buffer to be at the end
 		// of the out message. We need only shrink to the right size based on how
@@ -794,7 +997,7 @@ func (c *Connection) kernelResponseForOp(
 		out.St.Ffree = o.InodesFree
 		out.St.Namelen = 255
 
-		// The posix spec for sys/statvfs.h (http://goo.gl/LktgrF) defines the
+		// The posix spec for sys/statvfs.h (https://tinyurl.com/2juj6ah6) defines the
 		// following fields of statvfs, among others:
 		//
 		//     f_bsize    File system block size.
@@ -804,7 +1007,7 @@ func (c *Connection) kernelResponseForOp(
 		// It appears as though f_bsize was the only thing supported by most unixes
 		// originally, but then f_frsize was added when new sorts of file systems
 		// came about. Quoth The Linux Programming Interface by Michael Kerrisk
-		// (https://goo.gl/5LZMxQ):
+		// (https://tinyurl.com/5n8mjtws):
 		//
 		//     For most Linux file systems, the values of f_bsize and f_frsize are
 		//     the same. However, some file systems support the notion of block
@@ -848,6 +1051,9 @@ func (c *Connection) kernelResponseForOp(
 	case *fuseops.FallocateOp:
 		// Empty response
 
+	case *fuseops.SyncFSOp:
+		// Empty response
+
 	case *initOp:
 		out := (*fusekernel.InitOut)(m.Grow(int(unsafe.Sizeof(fusekernel.InitOut{}))))
 
@@ -862,6 +1068,13 @@ func (c *Connection) kernelResponseForOp(
 		out.TimeGran = 1
 		out.MaxPages = o.MaxPages
 
+	case *fuseops.PollOp:
+		out := (*fusekernel.PollOut)(m.Grow(int(unsafe.Sizeof(fusekernel.PollOut{}))))
+		out.Revents = uint32(o.Revents)
+
+	case *fuseops.NotifyRetrieveReplyOp:
+		// Empty response
+
 	default:
 		panic(fmt.Sprintf("Unexpected op: %#v", op))
 	}
@@ -869,124 +1082,73 @@ func (c *Connection) kernelResponseForOp(
 	return
 }
 
+// Like kernelResponse, but assumes the user replied with a nil error to the op.
+func (c *Connection) kernelNotification(
+	m *buffer.OutMessage,
+	op interface{}) {
+
+	h := m.OutHeader()
+	h.Unique = 0
+
+	// Create the appropriate output message
+	switch o := op.(type) {
+	case *fuseops.NotifyPollWakeup:
+		h.Error = fusekernel.NotifyCodePoll
+		out := (*fusekernel.NotifyPollWakeupOut)(m.Grow(int(unsafe.Sizeof(fusekernel.NotifyPollWakeupOut{}))))
+		out.Kh = o.Kh
+
+	case *fuseops.NotifyInvalInode:
+		h.Error = fusekernel.NotifyCodeInvalInode
+		out := (*fusekernel.NotifyInvalInodeOut)(m.Grow(int(unsafe.Sizeof(fusekernel.NotifyInvalInodeOut{}))))
+		out.Ino = uint64(o.Inode)
+		out.Off = o.Offset
+		out.Len = o.Length
+
+	case *fuseops.NotifyInvalEntry:
+		h.Error = fusekernel.NotifyCodeInvalEntry
+		out := (*fusekernel.NotifyInvalEntryOut)(m.Grow(int(unsafe.Sizeof(fusekernel.NotifyInvalEntryOut{}))))
+		out.Parent = uint64(o.Parent)
+		out.Namelen = uint32(len(o.Name))
+		m.AppendString(o.Name)
+		m.AppendString("\x00")
+
+	case *fuseops.NotifyDelete:
+		h.Error = fusekernel.NotifyCodeDelete
+		out := (*fusekernel.NotifyDeleteOut)(m.Grow(int(unsafe.Sizeof(fusekernel.NotifyDeleteOut{}))))
+		out.Parent = uint64(o.Parent)
+		out.Child = uint64(o.Child)
+		out.Namelen = uint32(len(o.Name))
+		m.AppendString(o.Name)
+		m.AppendString("\x00")
+
+	case *fuseops.NotifyStore:
+		h.Error = fusekernel.NotifyCodeStore
+		out := (*fusekernel.NotifyStoreOut)(m.Grow(int(unsafe.Sizeof(fusekernel.NotifyStoreOut{}))))
+		out.Nodeid = uint64(o.Inode)
+		out.Offset = o.Offset
+		out.Size = o.Length
+		m.Append(o.Data...)
+		m.ShrinkTo(buffer.OutMessageHeaderSize + int(unsafe.Sizeof(fusekernel.NotifyStoreOut{})) + int(o.Length))
+
+	case *fuseops.NotifyRetrieve:
+		h.Error = fusekernel.NotifyCodeRetrieve
+		out := (*fusekernel.NotifyRetrieveOut)(m.Grow(int(unsafe.Sizeof(fusekernel.NotifyRetrieveOut{}))))
+		out.Unique = o.Unique
+		out.Nodeid = uint64(o.Inode)
+		out.Offset = o.Offset
+		out.Size = o.Length
+
+	default:
+		panic(fmt.Sprintf("Unexpected notification: %#v", op))
+	}
+
+	return
+}
+
 ////////////////////////////////////////////////////////////////////////
 // General conversions
 ////////////////////////////////////////////////////////////////////////
 
-func convertTime(t time.Time) (secs uint64, nsec uint32) {
-	totalNano := t.UnixNano()
-	secs = uint64(totalNano / 1e9)
-	nsec = uint32(totalNano % 1e9)
-	return secs, nsec
-}
-
-func convertAttributes(
-	inodeID fuseops.InodeID,
-	in *fuseops.InodeAttributes,
-	out *fusekernel.Attr) {
-	out.Ino = uint64(inodeID)
-	out.Size = in.Size
-	out.Atime, out.AtimeNsec = convertTime(in.Atime)
-	out.Mtime, out.MtimeNsec = convertTime(in.Mtime)
-	out.Ctime, out.CtimeNsec = convertTime(in.Ctime)
-	out.SetCrtime(convertTime(in.Crtime))
-	out.Nlink = in.Nlink
-	out.Uid = in.Uid
-	out.Gid = in.Gid
-	// round up to the nearest 512 boundary
-	out.Blocks = (in.Size + 512 - 1) / 512
-
-	// Set the mode.
-	out.Mode = uint32(in.Mode) & 0777
-	switch {
-	default:
-		out.Mode |= syscall.S_IFREG
-	case in.Mode&os.ModeDir != 0:
-		out.Mode |= syscall.S_IFDIR
-	case in.Mode&os.ModeDevice != 0:
-		if in.Mode&os.ModeCharDevice != 0 {
-			out.Mode |= syscall.S_IFCHR
-		} else {
-			out.Mode |= syscall.S_IFBLK
-		}
-	case in.Mode&os.ModeNamedPipe != 0:
-		out.Mode |= syscall.S_IFIFO
-	case in.Mode&os.ModeSymlink != 0:
-		out.Mode |= syscall.S_IFLNK
-	case in.Mode&os.ModeSocket != 0:
-		out.Mode |= syscall.S_IFSOCK
-	}
-	if in.Mode&os.ModeSetuid != 0 {
-		out.Mode |= syscall.S_ISUID
-	}
-	if in.Mode&os.ModeSetgid != 0 {
-		out.Mode |= syscall.S_ISGID
-	}
-	if in.Mode&os.ModeSticky != 0 {
-		out.Mode |= syscall.S_ISVTX
-	}
-}
-
-// Convert an absolute cache expiration time to a relative time from now for
-// consumption by the fuse kernel module.
-func convertExpirationTime(t time.Time) (secs uint64, nsecs uint32) {
-	// Fuse represents durations as unsigned 64-bit counts of seconds and 32-bit
-	// counts of nanoseconds (cf. http://goo.gl/EJupJV). So negative durations
-	// are right out. There is no need to cap the positive magnitude, because
-	// 2^64 seconds is well longer than the 2^63 ns range of time.Duration.
-	d := t.Sub(time.Now())
-	if d > 0 {
-		secs = uint64(d / time.Second)
-		nsecs = uint32((d % time.Second) / time.Nanosecond)
-	}
-
-	return secs, nsecs
-}
-
-func convertChildInodeEntry(
-	in *fuseops.ChildInodeEntry,
-	out *fusekernel.EntryOut) {
-	out.Nodeid = uint64(in.Child)
-	out.Generation = uint64(in.Generation)
-	out.EntryValid, out.EntryValidNsec = convertExpirationTime(in.EntryExpiration)
-	out.AttrValid, out.AttrValidNsec = convertExpirationTime(in.AttributesExpiration)
-
-	convertAttributes(in.Child, &in.Attributes, &out.Attr)
-}
-
-func convertFileMode(unixMode uint32) os.FileMode {
-	mode := os.FileMode(unixMode & 0777)
-	switch unixMode & syscall.S_IFMT {
-	case syscall.S_IFREG:
-		// nothing
-	case syscall.S_IFDIR:
-		mode |= os.ModeDir
-	case syscall.S_IFCHR:
-		mode |= os.ModeCharDevice | os.ModeDevice
-	case syscall.S_IFBLK:
-		mode |= os.ModeDevice
-	case syscall.S_IFIFO:
-		mode |= os.ModeNamedPipe
-	case syscall.S_IFLNK:
-		mode |= os.ModeSymlink
-	case syscall.S_IFSOCK:
-		mode |= os.ModeSocket
-	default:
-		// no idea
-		mode |= os.ModeDevice
-	}
-	if unixMode&syscall.S_ISUID != 0 {
-		mode |= os.ModeSetuid
-	}
-	if unixMode&syscall.S_ISGID != 0 {
-		mode |= os.ModeSetgid
-	}
-	if unixMode&syscall.S_ISVTX != 0 {
-		mode |= os.ModeSticky
-	}
-	return mode
-}
-
 func writeXattrSize(m *buffer.OutMessage, size uint32) {
 	out := (*fusekernel.GetxattrOut)(m.Grow(int(unsafe.Sizeof(fusekernel.GetxattrOut{}))))
 	out.Size = size

debug.go

@@ -115,6 +115,9 @@ func describeRequest(op interface{}) (s string) {
 		addComponent("offset %d", typed.Offset)
 		addComponent("length %d", typed.Length)
 		addComponent("mode %d", typed.Mode)
+
+	case *fuseops.ReleaseFileHandleOp:
+		addComponent("handle %d", typed.Handle)
 	}
 
 	// Use just the name if there is no extra info.
@@ -141,6 +144,10 @@ func describeResponse(op interface{}) string {
 			addComponent("inode %v", entry.Child)
 		}
 	}
+	switch typed := op.(type) {
+	case *fuseops.OpenFileOp:
+		addComponent("handle %d", typed.Handle)
+	}
 
-	return fmt.Sprintf("%s", strings.Join(components, ", "))
+	return fmt.Sprintf("%s (%s)", opName(op), strings.Join(components, ", "))
 }

fuseops/conv.go

@@ -0,0 +1,85 @@
+// Copyright 2023 Vitaliy Filippov
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package fuseops
+
+import (
+	"time"
+	"syscall"
+
+	"github.com/jacobsa/fuse/internal/fusekernel"
+)
+
+////////////////////////////////////////////////////////////////////////
+// General conversions
+////////////////////////////////////////////////////////////////////////
+
+func ConvertTime(t time.Time) (secs uint64, nsec uint32) {
+	totalNano := t.UnixNano()
+	secs = uint64(totalNano / 1e9)
+	nsec = uint32(totalNano % 1e9)
+	return secs, nsec
+}
+
+func ConvertAttributes(
+	inodeID InodeID,
+	in *InodeAttributes,
+	out *fusekernel.Attr) {
+	out.Ino = uint64(inodeID)
+	out.Size = in.Size
+	out.Atime, out.AtimeNsec = ConvertTime(in.Atime)
+	out.Mtime, out.MtimeNsec = ConvertTime(in.Mtime)
+	out.Ctime, out.CtimeNsec = ConvertTime(in.Ctime)
+	out.SetCrtime(ConvertTime(in.Crtime))
+	out.Nlink = in.Nlink
+	out.Uid = in.Uid
+	out.Gid = in.Gid
+	// round up to the nearest 512 boundary
+	out.Blocks = (in.Size + 512 - 1) / 512
+
+	// Set the mode.
+	out.Mode = ConvertGoMode(in.Mode)
+
+	if out.Mode & (syscall.S_IFCHR | syscall.S_IFBLK) != 0 {
+		out.Rdev = in.Rdev
+	}
+}
+
+// Convert an absolute cache expiration time to a relative time from now for
+// consumption by the fuse kernel module.
+func ConvertExpirationTime(t time.Time) (secs uint64, nsecs uint32) {
+	// Fuse represents durations as unsigned 64-bit counts of seconds and 32-bit
+	// counts of nanoseconds (https://tinyurl.com/4muvkr6k). So negative
+	// durations are right out. There is no need to cap the positive magnitude,
+	// because 2^64 seconds is well longer than the 2^63 ns range of
+	// time.Duration.
+	d := t.Sub(time.Now())
+	if d > 0 {
+		secs = uint64(d / time.Second)
+		nsecs = uint32((d % time.Second) / time.Nanosecond)
+	}
+
+	return secs, nsecs
+}
+
+func ConvertChildInodeEntry(
+	in *ChildInodeEntry,
+	out *fusekernel.EntryOut) {
+	out.Nodeid = uint64(in.Child)
+	out.Generation = uint64(in.Generation)
+	out.EntryValid, out.EntryValidNsec = ConvertExpirationTime(in.EntryExpiration)
+	out.AttrValid, out.AttrValidNsec = ConvertExpirationTime(in.AttributesExpiration)
+
+	ConvertAttributes(in.Child, &in.Attributes, &out.Attr)
+}

fuseops/filemode.go

@@ -0,0 +1,74 @@
+package fuseops
+
+import (
+	"os"
+	"syscall"
+)
+
+// ConvertFileMode returns an os.FileMode with the Go mode and permission bits
+// set according to the Linux mode and permission bits.
+func ConvertFileMode(unixMode uint32) os.FileMode {
+	mode := os.FileMode(unixMode & 0777)
+	switch unixMode & syscall.S_IFMT {
+	case syscall.S_IFREG:
+		// nothing
+	case syscall.S_IFDIR:
+		mode |= os.ModeDir
+	case syscall.S_IFCHR:
+		mode |= os.ModeCharDevice | os.ModeDevice
+	case syscall.S_IFBLK:
+		mode |= os.ModeDevice
+	case syscall.S_IFIFO:
+		mode |= os.ModeNamedPipe
+	case syscall.S_IFLNK:
+		mode |= os.ModeSymlink
+	case syscall.S_IFSOCK:
+		mode |= os.ModeSocket
+	default:
+		// no idea
+	}
+	if unixMode&syscall.S_ISUID != 0 {
+		mode |= os.ModeSetuid
+	}
+	if unixMode&syscall.S_ISGID != 0 {
+		mode |= os.ModeSetgid
+	}
+	if unixMode&syscall.S_ISVTX != 0 {
+		mode |= os.ModeSticky
+	}
+	return mode
+}
+
+// ConvertGoMode returns an integer with the Linux mode and permission bits
+// set according to the Go mode and permission bits.
+func ConvertGoMode(inMode os.FileMode) uint32 {
+	outMode := uint32(inMode) & 0777
+	switch {
+	default:
+		outMode |= syscall.S_IFREG
+	case inMode&os.ModeDir != 0:
+		outMode |= syscall.S_IFDIR
+	case inMode&os.ModeDevice != 0:
+		if inMode&os.ModeCharDevice != 0 {
+			outMode |= syscall.S_IFCHR
+		} else {
+			outMode |= syscall.S_IFBLK
+		}
+	case inMode&os.ModeNamedPipe != 0:
+		outMode |= syscall.S_IFIFO
+	case inMode&os.ModeSymlink != 0:
+		outMode |= syscall.S_IFLNK
+	case inMode&os.ModeSocket != 0:
+		outMode |= syscall.S_IFSOCK
+	}
+	if inMode&os.ModeSetuid != 0 {
+		outMode |= syscall.S_ISUID
+	}
+	if inMode&os.ModeSetgid != 0 {
+		outMode |= syscall.S_ISGID
+	}
+	if inMode&os.ModeSticky != 0 {
+		outMode |= syscall.S_ISVTX
+	}
+	return outMode
+}

fuseops/ops.go

@@ -28,25 +28,35 @@ import (
 // OpContext contains extra context that may be needed by some file systems.
 // See https://libfuse.github.io/doxygen/structfuse__context.html as a reference.
 type OpContext struct {
+	// FuseID is the Unique identifier for each operation from the kernel.
+	FuseID uint64
+
 	// PID of the process that is invoking the operation.
 	// Not filled in case of a writepage operation.
 	Pid uint32
+
+	// UID of the process that is invoking the operation.
+	// Not filled in case of a writepage operation.
+	Uid uint32
+
+	// GID of the process that is invoking the operation.
+	// Not filled in case of a writepage operation.
+	Gid uint32
 }
 
 // Return statistics about the file system's capacity and available resources.
 //
 // Called by statfs(2) and friends:
 //
-//   - (https://goo.gl/Xi1lDr) sys_statfs called user_statfs, which calls
+//   - (https://tinyurl.com/234ppacj) sys_statfs called user_statfs, which calls
 //     vfs_statfs, which calls statfs_by_dentry.
 //
-//   - (https://goo.gl/VAIOwU) statfs_by_dentry calls the superblock
+//   - (https://tinyurl.com/u6keadjz) statfs_by_dentry calls the superblock
 //     operation statfs, which in our case points at
-//     fuse_statfs (cf. https://goo.gl/L7BTM3)
+//     fuse_statfs (https://tinyurl.com/mr45wd28)
 //
-//   - (https://goo.gl/Zn7Sgl) fuse_statfs sends a statfs op, then uses
-//     convert_fuse_statfs to convert the response in a straightforward
-//     manner.
+//   - (https://tinyurl.com/3wt3dw3c) fuse_statfs sends a statfs op, then uses
+//     convert_fuse_statfs to convert the response in a straightforward manner.
 //
 // This op is particularly important on OS X: if you don't implement it, the
 // file system will not successfully mount. If you don't model a sane amount of
@@ -57,15 +67,15 @@ type StatFSOp struct {
 	// system's capacity and space availability.
 	//
 	// On Linux this is surfaced as statfs::f_frsize, matching the posix standard
-	// (http://goo.gl/LktgrF), which says that f_blocks and friends are in units
-	// of f_frsize. On OS X this is surfaced as statfs::f_bsize, which plays the
-	// same roll.
+	// (https://tinyurl.com/2juj6ah6), which says that f_blocks and friends are
+	// in units of f_frsize. On OS X this is surfaced as statfs::f_bsize, which
+	// plays the same roll.
 	//
 	// It appears as though the original intent of statvfs::f_frsize in the posix
 	// standard was to support a smaller addressable unit than statvfs::f_bsize
 	// (cf. The Linux Programming Interface by Michael Kerrisk,
-	// https://goo.gl/5LZMxQ). Therefore users should probably arrange for this
-	// to be no larger than IoSize.
+	// https://tinyurl.com/5n8mjtws). Therefore users should probably arrange for
+	// this to be no larger than IoSize.
 	//
 	// On Linux this can be any value, and will be faithfully returned to the
 	// caller of statfs(2) (see the code walk above). On OS X it appears that
@@ -183,20 +193,22 @@ type SetInodeAttributesOp struct {
 // contain a note of this (but see also the note about the root inode below).
 // For example, LookUpInodeOp and MkDirOp. The authoritative source is the
 // libfuse documentation, which states that any op that returns
-// fuse_reply_entry fuse_reply_create implicitly increments (cf.
-// http://goo.gl/o5C7Dx).
+// fuse_reply_entry fuse_reply_create implicitly increments
+// (https://tinyurl.com/2xd5zssm).
 //
 // If the reference count hits zero, the file system can forget about that ID
 // entirely, and even re-use it in future responses. The kernel guarantees that
 // it will not otherwise use it again.
 //
 // The reference count corresponds to fuse_inode::nlookup
-// (http://goo.gl/ut48S4). Some examples of where the kernel manipulates it:
+// (https://tinyurl.com/ycka69ck). Some examples of where the kernel
+// manipulates it:
 //
-//   - (http://goo.gl/vPD9Oh) Any caller to fuse_iget increases the count.
-//   - (http://goo.gl/B6tTTC) fuse_lookup_name calls fuse_iget.
-//   - (http://goo.gl/IlcxWv) fuse_create_open calls fuse_iget.
-//   - (http://goo.gl/VQMQul) fuse_dentry_revalidate increments after
+//   - (https://tinyurl.com/s8dz2ays) Any caller to fuse_iget increases the
+//     count.
+//   - (https://tinyurl.com/mu37ceua) fuse_lookup_name calls fuse_iget.
+//   - (https://tinyurl.com/2nyhhnsh) fuse_create_open calls fuse_iget.
+//   - (https://tinyurl.com/mnjpu3a9) fuse_dentry_revalidate increments after
 //     revalidating.
 //
 // In contrast to all other inodes, RootInodeID begins with an implicit
@@ -204,12 +216,13 @@ type SetInodeAttributesOp struct {
 // could be no such op, because the root cannot be referred to by name.) Code
 // walk:
 //
-//   - (http://goo.gl/gWAheU) fuse_fill_super calls fuse_get_root_inode.
+//   - (https://tinyurl.com/yf8m2drx) fuse_fill_super calls
+//     fuse_get_root_inode.
 //
-//   - (http://goo.gl/AoLsbb) fuse_get_root_inode calls fuse_iget without
-//     sending any particular request.
+//   - (https://tinyurl.com/35f86asu) fuse_get_root_inode calls fuse_iget
+//     without sending any particular request.
 //
-//   - (http://goo.gl/vPD9Oh) fuse_iget increments nlookup.
+//   - (https://tinyurl.com/s8dz2ays) fuse_iget increments nlookup.
 //
 // File systems should tolerate but not rely on receiving forget ops for
 // remaining inodes when the file system unmounts, including the root inode.
@@ -259,10 +272,11 @@ type BatchForgetOp struct {
 //
 // The Linux kernel appears to verify the name doesn't already exist (mkdir
 // calls mkdirat calls user_path_create calls filename_create, which verifies:
-// http://goo.gl/FZpLu5). Indeed, the tests in samples/memfs that call in
-// parallel appear to bear this out. But osxfuse does not appear to guarantee
-// this (cf. https://goo.gl/PqzZDv). And if names may be created outside of the
-// kernel's control, it doesn't matter what the kernel does anyway.
+// https://tinyurl.com/24yw46mf). Indeed, the tests in samples/memfs that call
+// in parallel appear to bear this out. But osxfuse does not appear to
+// guarantee this (https://tinyurl.com/22587hcf). And if names may be created
+// outside of the kernel's control, it doesn't matter what the kernel does
+// anyway.
 //
 // Therefore the file system should return EEXIST if the name already exists.
 type MkDirOp struct {
@@ -283,14 +297,14 @@ type MkDirOp struct {
 
 // Create a file inode as a child of an existing directory inode. The kernel
 // sends this in response to a mknod(2) call. It may also send it in special
-// cases such as an NFS export (cf. https://goo.gl/HiLfnK). It is more typical
+// cases such as an NFS export (https://tinyurl.com/5dwxr7c9). It is more typical
 // to see CreateFileOp, which is received for an open(2) that creates a file.
 //
 // The Linux kernel appears to verify the name doesn't already exist (mknod
 // calls sys_mknodat calls user_path_create calls filename_create, which
-// verifies: http://goo.gl/FZpLu5). But osxfuse may not guarantee this, as with
-// mkdir(2). And if names may be created outside of the kernel's control, it
-// doesn't matter what the kernel does anyway.
+// verifies: https://tinyurl.com/24yw46mf). But osxfuse may not guarantee this,
+// as with mkdir(2). And if names may be created outside of the kernel's
+// control, it doesn't matter what the kernel does anyway.
 //
 // Therefore the file system should return EEXIST if the name already exists.
 type MkNodeOp struct {
@@ -301,6 +315,9 @@ type MkNodeOp struct {
 	Name string
 	Mode os.FileMode
 
+	// The device number (only valid if created file is a device)
+	Rdev uint32
+
 	// Set by the file system: information about the inode that was created.
 	//
 	// The lookup count for the inode is implicitly incremented. See notes on
@@ -313,10 +330,10 @@ type MkNodeOp struct {
 //
 // The kernel sends this when the user asks to open a file with the O_CREAT
 // flag and the kernel has observed that the file doesn't exist. (See for
-// example lookup_open, http://goo.gl/PlqE9d). However, osxfuse doesn't appear
-// to make this check atomically (cf. https://goo.gl/PqzZDv). And if names may
-// be created outside of the kernel's control, it doesn't matter what the
-// kernel does anyway.
+// example lookup_open, https://tinyurl.com/49899mvb). However, osxfuse doesn't
+// appear to make this check atomically (https://tinyurl.com/22587hcf). And if
+// names may be created outside of the kernel's control, it doesn't matter what
+// the kernel does anyway.
 //
 // Therefore the file system should return EEXIST if the name already exists.
 type CreateFileOp struct {
@@ -394,8 +411,8 @@ type CreateLinkOp struct {
 // Rename a file or directory, given the IDs of the original parent directory
 // and the new one (which may be the same).
 //
-// In Linux, this is called by vfs_rename (https://goo.gl/eERItT), which is
-// called by sys_renameat2 (https://goo.gl/fCC9qC).
+// In Linux, this is called by vfs_rename (https://tinyurl.com/2xbx9kr2), which
+// is called by sys_renameat2 (https://tinyurl.com/4zyak2kt).
 //
 // The kernel takes care of ensuring that the source and destination are not
 // identical (in which case it does nothing), that the rename is not across
@@ -404,7 +421,7 @@ type CreateLinkOp struct {
 //
 //   - If the new name is an existing directory, the file system must ensure it
 //     is empty before replacing it, returning ENOTEMPTY otherwise. (This is
-//     per the posix spec: http://goo.gl/4XtT79)
+//     per the posix spec: https://tinyurl.com/5n865nx9)
 //
 //   - The rename must be atomic from the point of view of an observer of the
 //     new name. That is, if the new name already exists, there must be no
@@ -412,9 +429,9 @@ type CreateLinkOp struct {
 //
 //   - It is okay for the new name to be modified before the old name is
 //     removed; these need not be atomic. In fact, the Linux man page
-//     explicitly says this is likely (cf. https://goo.gl/Y1wVZc).
+//     explicitly says this is likely (https://tinyurl.com/mdpbpjmr).
 //
-//   - Linux bends over backwards (https://goo.gl/pLDn3r) to ensure that
+//   - Linux bends over backwards (https://tinyurl.com/3hmt7puy) to ensure that
 //     neither the old nor the new parent can be concurrently modified. But
 //     it's not clear whether OS X does this, and in any case it doesn't matter
 //     for file systems that may be modified remotely. Therefore a careful file
@@ -443,7 +460,7 @@ type RenameOp struct {
 //
 // The file system is responsible for checking that the directory is empty.
 //
-// Sample implementation in ext2: ext2_rmdir (http://goo.gl/B9QmFf)
+// Sample implementation in ext2: ext2_rmdir (https://tinyurl.com/bajkpcf9)
 type RmDirOp struct {
 	// The ID of parent directory inode, and the name of the directory being
 	// removed within it.
@@ -457,7 +474,7 @@ type RmDirOp struct {
 // ForgetInodeOp. It may still be referenced before then if a user still has
 // the file open.
 //
-// Sample implementation in ext2: ext2_unlink (http://goo.gl/hY6r6C)
+// Sample implementation in ext2: ext2_unlink (https://tinyurl.com/3wpwedcp)
 type UnlinkOp struct {
 	// The ID of parent directory inode, and the name of the entry being removed
 	// within it.
@@ -490,6 +507,14 @@ type OpenDirOp struct {
 	// a later call to ReleaseDirHandle.
 	Handle    HandleID
 	OpContext OpContext
+
+	// CacheDir conveys to the kernel to cache the response of next
+	// ReadDirOp as page cache. Once cached, listing on that directory will be
+	// served from the kernel until invalidated.
+	CacheDir bool
+
+	// KeepCache instructs the kernel to not invalidate the data cache on open calls.
+	KeepCache bool
 }
 
 // Read entries from a directory previously opened with OpenDir.
@@ -509,47 +534,51 @@ type ReadDirOp struct {
 	// at zero and is set by llseek and by the final consumed result returned by
 	// each call to ReadDir:
 	//
-	//  *  (http://goo.gl/2nWJPL) iterate_dir, which is called by getdents(2) and
-	//     readdir(2), sets dir_context::pos to file::f_pos before calling
-	//     f_op->iterate, and then does the opposite assignment afterward.
+	//  *  (https://tinyurl.com/3ueykmaj) iterate_dir, which is called by
+	//     getdents(2) and readdir(2), sets dir_context::pos to file::f_pos
+	//     before calling f_op->iterate, and then does the opposite assignment
+	//     afterward.
 	//
-	//  *  (http://goo.gl/rTQVSL) fuse_readdir, which implements iterate for fuse
-	//     directories, passes dir_context::pos as the offset to fuse_read_fill,
-	//     which passes it on to user-space. fuse_readdir later calls
-	//     parse_dirfile with the same context.
+	//  *  (https://tinyurl.com/a8urhfy9) fuse_readdir, which implements iterate
+	//     for fuse directories, passes dir_context::pos as the offset to
+	//     fuse_read_fill, which passes it on to user-space. fuse_readdir later
+	//     calls parse_dirfile with the same context.
 	//
-	//  *  (http://goo.gl/vU5ukv) For each returned result (except perhaps the
-	//     last, which may be truncated by the page boundary), parse_dirfile
-	//     updates dir_context::pos with fuse_dirent::off.
+	//  *  (https://tinyurl.com/5cev5fn4) For each returned result (except
+	//     perhaps the last, which may be truncated by the page boundary),
+	//     parse_dirfile updates dir_context::pos with fuse_dirent::off.
 	//
 	// It is affected by the Posix directory stream interfaces in the following
 	// manner:
 	//
-	//  *  (http://goo.gl/fQhbyn, http://goo.gl/ns1kDF) opendir initially causes
-	//     filepos to be set to zero.
+	//  *  (https://tinyurl.com/2pjv5jvz, https://tinyurl.com/2r6h4mkj) opendir
+	//     initially causes filepos to be set to zero.
 	//
-	//  *  (http://goo.gl/ezNKyR, http://goo.gl/xOmDv0) readdir allows the user
-	//     to iterate through the directory one entry at a time. As each entry is
-	//     consumed, its d_off field is stored in __dirstream::filepos.
+	//  *  (https://tinyurl.com/2yvcbcpv, https://tinyurl.com/bddezwp4) readdir
+	//     allows the user to iterate through the directory one entry at a time.
+	//     As each entry is consumed, its d_off field is stored in
+	//     __dirstream::filepos.
 	//
-	//  *  (http://goo.gl/WEOXG8, http://goo.gl/rjSXl3) telldir allows the user
-	//     to obtain the d_off field from the most recently returned entry.
+	//  *  (https://tinyurl.com/2pfbfe9v, https://tinyurl.com/4wtat58a) telldir
+	//     allows the user to obtain the d_off field from the most recently
+	//     returned entry.
 	//
-	//  *  (http://goo.gl/WG3nDZ, http://goo.gl/Lp0U6W) seekdir allows the user
-	//     to seek backward to an offset previously returned by telldir. It
-	//     stores the new offset in filepos, and calls llseek to update the
-	//     kernel's struct file.
+	//  *  (https://tinyurl.com/bdynryef, https://tinyurl.com/4hysrnb8) seekdir
+	//     allows the user to seek backward to an offset previously returned by
+	//     telldir. It stores the new offset in filepos, and calls llseek to
+	//     update the kernel's struct file.
 	//
-	//  *  (http://goo.gl/gONQhz, http://goo.gl/VlrQkc) rewinddir allows the user
-	//     to go back to the beginning of the directory, obtaining a fresh view.
-	//     It updates filepos and calls llseek to update the kernel's struct
-	//     file.
+	//  *  (https://tinyurl.com/5n8dkb44, https://tinyurl.com/3jnn5nnn) rewinddir
+	//     allows the user to go back to the beginning of the directory,
+	//     obtaining a fresh view. It updates filepos and calls llseek to update
+	//     the kernel's struct file.
 	//
 	// Unfortunately, FUSE offers no way to intercept seeks
-	// (http://goo.gl/H6gEXa), so there is no way to cause seekdir or rewinddir
-	// to fail. Additionally, there is no way to distinguish an explicit
-	// rewinddir followed by readdir from the initial readdir, or a rewinddir
-	// from a seekdir to the value returned by telldir just after opendir.
+	// (https://tinyurl.com/4bm2sfjd), so there is no way to cause seekdir or
+	// rewinddir to fail. Additionally, there is no way to distinguish an
+	// explicit rewinddir followed by readdir from the initial readdir, or a
+	// rewinddir from a seekdir to the value returned by telldir just after
+	// opendir.
 	//
 	// Luckily, Posix is vague about what the user will see if they seek
 	// backwards, and requires the user not to seek to an old offset after a
@@ -559,12 +588,18 @@ type ReadDirOp struct {
 	// offset, and return array offsets into that cached listing.
 	Offset DirOffset
 
+	// Whether this operation is a READDIRPLUS
+	//
+	// If true, then the FS must return inode attributes and expiration time
+	// along with each directory entry and increment its reference count.
+	Plus bool
+
 	// The destination buffer, whose length gives the size of the read.
 	//
 	// The output data should consist of a sequence of FUSE directory entries in
-	// the format generated by fuse_add_direntry (http://goo.gl/qCcHCV), which is
-	// consumed by parse_dirfile (http://goo.gl/2WUmD2). Use fuseutil.WriteDirent
-	// to generate this data.
+	// the format generated by fuse_add_direntry (https://tinyurl.com/3r9t7d2p),
+	// which is consumed by parse_dirfile (https://tinyurl.com/bevwty74). Use
+	// fuseutil.WriteDirent or fuseutil.WriteDirentPlus to generate this data.
 	//
 	// Each entry returned exposes a directory offset to the user that may later
 	// show up in ReadDirRequest.Offset. See notes on that field for more
@@ -577,10 +612,10 @@ type ReadDirOp struct {
 	// entries available or the final entry would not fit.
 	//
 	// Zero means that the end of the directory has been reached. This is
-	// unambiguous because NAME_MAX (https://goo.gl/ZxzKaE) plus the size of
-	// fuse_dirent (https://goo.gl/WO8s3F) plus the 8-byte alignment of
-	// FUSE_DIRENT_ALIGN (http://goo.gl/UziWvH) is less than the read size of
-	// PAGE_SIZE used by fuse_readdir (cf. https://goo.gl/VajtS2).
+	// unambiguous because NAME_MAX (https://tinyurl.com/4r2b68jp) plus the size
+	// of fuse_dirent (https://tinyurl.com/mp43bu8) plus the 8-byte alignment of
+	// FUSE_DIRENT_ALIGN (https://tinyurl.com/3m3ewu7h) is less than the read
+	// size of PAGE_SIZE used by fuse_readdir (https://tinyurl.com/mrwxsfxw).
 	BytesRead int
 	OpContext OpContext
 }
@@ -592,7 +627,8 @@ type ReadDirOp struct {
 // The kernel guarantees that the handle ID will not be used in further ops
 // sent to the file system (unless it is reissued by the file system).
 //
-// Errors from this op are ignored by the kernel (cf. http://goo.gl/RL38Do).
+// Errors from this op are ignored by the kernel
+// (https://tinyurl.com/2aaccyzk).
 type ReleaseDirHandleOp struct {
 	// The handle ID to be released. The kernel guarantees that this ID will not
 	// be used in further calls to the file system (unless it is reissued by the
@@ -625,15 +661,15 @@ type OpenFileOp struct {
 	Handle HandleID
 
 	// By default, fuse invalidates the kernel's page cache for an inode when a
-	// new file handle is opened for that inode (cf. https://goo.gl/2rZ9uk). The
-	// intent appears to be to allow users to "see" content that has changed
+	// new file handle is opened for that inode (https://tinyurl.com/yyb497zy).
+	// The intent appears to be to allow users to "see" content that has changed
 	// remotely on a networked file system by re-opening the file.
 	//
 	// For file systems where this is not a concern because all modifications for
 	// a particular inode go through the kernel, set this field to true to
 	// disable this behavior.
 	//
-	// (More discussion: http://goo.gl/cafzWF)
+	// (More discussion: https://tinyurl.com/4znxvzwh)
 	//
 	// Note that on OS X it appears that the behavior is always as if this field
 	// is set to true, regardless of its value, at least for files opened in the
@@ -682,15 +718,21 @@ type ReadFileOp struct {
 	// Set by the file system: the number of bytes read.
 	//
 	// The FUSE documentation requires that exactly the requested number of bytes
-	// be returned, except in the case of EOF or error (http://goo.gl/ZgfBkF).
-	// This appears to be because it uses file mmapping machinery
-	// (http://goo.gl/SGxnaN) to read a page at a time. It appears to understand
-	// where EOF is by checking the inode size (http://goo.gl/0BkqKD), returned
-	// by a previous call to LookUpInode, GetInodeAttributes, etc.
+	// be returned, except in the case of EOF or error
+	// (https://tinyurl.com/2mzewn35). This appears to be because it uses file
+	// mmapping machinery (https://tinyurl.com/avxy3dvm) to read a page at a
+	// time. It appears to understand where EOF is by checking the inode size
+	// (https://tinyurl.com/2eteerzt), returned by a previous call to
+	// LookUpInode, GetInodeAttributes, etc.
 	//
 	// If direct IO is enabled, semantics should match those of read(2).
 	BytesRead int
 	OpContext OpContext
+
+	// If set, this function will be invoked after the operation response has been
+	// sent to the kernel and before the buffers containing the response data are
+	// freed.
+	Callback func()
 }
 
 // Write data to a file previously opened with CreateFile or OpenFile.
@@ -700,21 +742,22 @@ type ReadFileOp struct {
 // page via the FUSE VFS layer, causing this op to be sent:
 //
 //   - The kernel calls address_space_operations::writepage when a dirty page
-//     needs to be written to backing store (cf. http://goo.gl/Ezbewg). Fuse
-//     sets this to fuse_writepage (cf. http://goo.gl/IeNvLT).
+//     needs to be written to backing store (https://tinyurl.com/yck2sf5u).
+//     Fuse sets this to fuse_writepage (https://tinyurl.com/5n989f8p).
 //
-//   - (http://goo.gl/Eestuy) fuse_writepage calls fuse_writepage_locked.
+//   - (https://tinyurl.com/mvn6zv3j) fuse_writepage calls
+//     fuse_writepage_locked.
 //
-//   - (http://goo.gl/RqYIxY) fuse_writepage_locked makes a write request to
-//     the userspace server.
+//   - (https://tinyurl.com/2wn8scwb) fuse_writepage_locked makes a write
+//     request to the userspace server.
 //
 // Note that the kernel *will* ensure that writes are received and acknowledged
 // by the file system before sending a FlushFileOp when closing the file
 // descriptor to which they were written. Cf. the notes on
 // fuse.MountConfig.DisableWritebackCaching.
 //
-// (See also http://goo.gl/ocdTdM, fuse-devel thread "Fuse guarantees on
-// concurrent requests".)
+// (See also https://tinyurl.com/5dchkdtx, fuse-devel thread "Fuse guarantees
+// on concurrent requests".)
 type WriteFileOp struct {
 	// The file inode that we are modifying, and the handle previously returned
 	// by CreateFile or OpenFile when opening that inode.
@@ -746,25 +789,41 @@ type WriteFileOp struct {
 	// The data to write.
 	//
 	// The FUSE documentation requires that exactly the number of bytes supplied
-	// be written, except on error (http://goo.gl/KUpwwn). This appears to be
-	// because it uses file mmapping machinery (http://goo.gl/SGxnaN) to write a
-	// page at a time.
+	// be written, except on error (https://tinyurl.com/yuruk5tx). This appears
+	// to be because it uses file mmapping machinery
+	// (https://tinyurl.com/avxy3dvm) to write a page at a time.
 	Data      []byte
+
+	// Set by the file system: "no reuse" flag.
+	//
+	// By default, the Data buffer is reused by the library, so the file system
+	// must copy the data if it wants to use it later.
+	//
+	// However, if the file system sets this flag to true, the library doesn't
+	// reuse this buffer, so the file system can safely store and use Data slice
+	// without copying memory.
+	SuppressReuse bool
+
 	OpContext OpContext
+
+	// If set, this function will be invoked after the operation response has been
+	// sent to the kernel and before the buffers containing the response data are
+	// freed.
+	Callback func()
 }
 
 // Synchronize the current contents of an open file to storage.
 //
 // vfs.txt documents this as being called for by the fsync(2) system call
-// (cf. http://goo.gl/j9X8nB). Code walk for that case:
+// (https://tinyurl.com/y2kdrfzw). Code walk for that case:
 //
-//   - (http://goo.gl/IQkWZa) sys_fsync calls do_fsync, calls vfs_fsync, calls
-//     vfs_fsync_range.
+//   - (https://tinyurl.com/2s44cefz) sys_fsync calls do_fsync, calls
+//     vfs_fsync, calls vfs_fsync_range.
 //
-//   - (http://goo.gl/5L2SMy) vfs_fsync_range calls f_op->fsync.
+//   - (https://tinyurl.com/bdhhfam5) vfs_fsync_range calls f_op->fsync.
 //
-// Note that this is also sent by fdatasync(2) (cf. http://goo.gl/01R7rF), and
-// may be sent for msync(2) with the MS_SYNC flag (see the notes on
+// Note that this is also sent by fdatasync(2) (https://tinyurl.com/ja5wtszf),
+// and may be sent for msync(2) with the MS_SYNC flag (see the notes on
 // FlushFileOp).
 //
 // See also: FlushFileOp, which may perform a similar function when closing a
@@ -779,39 +838,42 @@ type SyncFileOp struct {
 // Flush the current state of an open file to storage upon closing a file
 // descriptor.
 //
-// vfs.txt documents this as being sent for each close(2) system call (cf.
-// http://goo.gl/FSkbrq). Code walk for that case:
+// vfs.txt documents this as being sent for each close(2) system call
+// (https://tinyurl.com/r4ujfxkc). Code walk for that case:
 //
-//   - (http://goo.gl/e3lv0e) sys_close calls __close_fd, calls filp_close.
-//   - (http://goo.gl/nI8fxD) filp_close calls f_op->flush (fuse_flush).
+//   - (https://tinyurl.com/2kzyyjcu) sys_close calls __close_fd, calls
+//     filp_close.
+
+//   - (https://tinyurl.com/4zdxrz52) filp_close calls f_op->flush
+//     (fuse_flush).
 //
 // But note that this is also sent in other contexts where a file descriptor is
-// closed, such as dup2(2) (cf. http://goo.gl/NQDvFS). In the case of close(2),
-// a flush error is returned to the user. For dup2(2), it is not.
+// closed, such as dup2(2) (https://tinyurl.com/5bj3z3f5). In the case of
+// close(2), a flush error is returned to the user. For dup2(2), it is not.
 //
 // One potentially significant case where this may not be sent is mmap'd files,
 // where the behavior is complicated:
 //
-//   - munmap(2) does not cause flushes (cf. http://goo.gl/j8B9g0).
+//   - munmap(2) does not cause flushes (https://tinyurl.com/ycy9z2jb).
 //
-//   - On OS X, if a user modifies a mapped file via the mapping before
-//     closing the file with close(2), the WriteFileOps for the modifications
-//     may not be received before the FlushFileOp for the close(2) (cf.
+//   - On OS X, if a user modifies a mapped file via the mapping before closing
+//     the file with close(2), the WriteFileOps for the modifications may not
+//     be received before the FlushFileOp for the close(2) (cf.
 //     https://github.com/osxfuse/osxfuse/issues/202). It appears that this may
-//     be fixed in osxfuse 3 (cf. https://goo.gl/rtvbko).
+//     be fixed in osxfuse 3 (https://tinyurl.com/2ne2jv8u).
 //
-//   - However, you safely can arrange for writes via a mapping to be
-//     flushed by calling msync(2) followed by close(2). On OS X msync(2)
-//     will cause a WriteFileOps to go through and close(2) will cause a
-//     FlushFile as usual (cf. http://goo.gl/kVmNcx). On Linux, msync(2) does
-//     nothing unless you set the MS_SYNC flag, in which case it causes a
-//     SyncFileOp to be sent (cf. http://goo.gl/P3mErk).
+//   - However, you safely can arrange for writes via a mapping to be flushed
+//     by calling msync(2) followed by close(2). On OS X msync(2) will cause a
+//     WriteFileOps to go through and close(2) will cause a FlushFile as usual
+//     (https://tinyurl.com/2p9b4axf). On Linux, msync(2) does nothing unless
+//     you set the MS_SYNC flag, in which case it causes a SyncFileOp to be
+//     sent (https://tinyurl.com/2y3d9hhj).
 //
 // In summary: if you make data durable in both FlushFile and SyncFile, then
 // your users can get safe behavior from mapped files on both operating systems
 // by calling msync(2) with MS_SYNC, followed by munmap(2), followed by
-// close(2). On Linux, the msync(2) is optional (cf. http://goo.gl/EIhAxv and
-// the notes on WriteFileOp).
+// close(2). On Linux, the msync(2) is optional (cf.
+// https://tinyurl.com/unesszdp and the notes on WriteFileOp).
 //
 // Because of cases like dup2(2), FlushFileOps are not necessarily one to one
 // with OpenFileOps. They should not be used for reference counting, and the
@@ -838,7 +900,8 @@ type FlushFileOp struct {
 // The kernel guarantees that the handle ID will not be used in further calls
 // to the file system (unless it is reissued by the file system).
 //
-// Errors from this op are ignored by the kernel (cf. http://goo.gl/RL38Do).
+// Errors from this op are ignored by the kernel
+// (https://tinyurl.com/2aaccyzk).
 type ReleaseFileHandleOp struct {
 	// The handle ID to be released. The kernel guarantees that this ID will not
 	// be used in further calls to the file system (unless it is reissued by the
@@ -962,3 +1025,132 @@ type FallocateOp struct {
 	Mode      uint32
 	OpContext OpContext
 }
+
+type SyncFSOp struct {
+	Inode     InodeID
+	OpContext OpContext
+}
+
+// Request notifications when the file system user calls poll/select or
+// similar operations on a file.
+type PollOp struct {
+	// The inode and handle the user wants to poll
+	Inode  InodeID
+	Handle HandleID
+
+	// Kh is the "kernel handle". The reason behind it is that it's allocated
+	// by the kernel on file allocation and guaranteed to be unique as opposed
+	// to regular file handles (HandleID) generated by the userland server
+	// (by us). Kh has to be used in NotifyPollWakeupOut replies.
+	Kh uint64
+
+	// Poll flags
+	Flags fusekernel.PollFlags
+
+	// Requested events
+	Events fusekernel.PollEvents
+
+	// Set by the file system: the actual events that have happened
+	// since the last poll
+	Revents   fusekernel.PollEvents
+	OpContext OpContext
+}
+
+// Notify consumers waiting for poll/epoll that events are incoming
+// for the specified kernel handle. The kernel will send a PollOp request
+// to get the event mask after receiving this notification
+type NotifyPollWakeup struct {
+	Kh uint64
+}
+
+// Notify to invalidate cache for an inode.
+//
+// If the filesystem has writeback caching enabled, invalidating an inode
+// will first trigger a writeback of all dirty pages. The call will block
+// until all writeback requests have completed and the inode has been
+// invalidated. It will, however, not wait for completion of pending writeback
+// requests that have been issued before.
+type NotifyInvalInode struct {
+	Inode  InodeID
+	Offset int64
+	Length int64
+}
+
+// Notify to invalidate parent attributes and the dentry matching parent/name
+//
+// To avoid a deadlock this request must not be sent in the execution path
+// of a related filesytem operation or within any code that could hold a lock
+// that could be needed to execute such an operation. As of kernel 4.18, a
+// "related operation" is a lookup(), symlink(), mknod(), mkdir(), unlink(),
+// rename(), link() or create() request for the parent, and a setattr(),
+// unlink(), rmdir(), rename(), setxattr(), removexattr(), readdir() or
+// readdirplus() request for the inode itself.
+//
+// When called correctly, it will never block.
+type NotifyInvalEntry struct {
+	Parent InodeID
+	Name   string
+}
+
+// This request behaves like NotifyInvalEntry with the following additional
+// effect (at least as of Linux kernel 4.8):
+//
+// If the provided child inode matches the inode that is currently associated
+// with the cached dentry, and if there are any inotify watches registered for
+// the dentry, then the watchers are informed that the dentry has been deleted.
+//
+// To avoid a deadlock this request must not be sent while executing a
+// related filesytem operation or while holding a lock that could be needed to
+// execute such an operation.
+type NotifyDelete struct {
+	Parent InodeID
+	Child  InodeID
+	Name   string
+}
+
+// Store data to the kernel buffers
+//
+// Synchronously store data in the kernel buffers belonging to the given inode.
+// The stored data is marked up-to-date (no read will be performed against it,
+// unless it's invalidated or evicted from the cache).
+//
+// If the stored data overflows the current file size, then the size is extended,
+// similarly to a write(2) on the filesystem.
+//
+// If this request returns an error, then the store wasn't fully completed, but
+// it may have been partially completed.
+type NotifyStore struct {
+	Inode  InodeID
+	Offset uint64
+	Length uint32
+	Data   [][]byte
+}
+
+// Retrieve data from the kernel buffers belonging to the given inode
+//
+// If successful then the kernel will send a NotifyRetrieveReplyOp as a reply.
+// Only present pages are returned in the retrieve reply. Retrieving stops when it
+// finds a non-present page and only data prior to that is returned.
+//
+// If this request returns an error, then the retrieve will not be completed and
+// no reply will be sent.
+//
+// This request doesn't change the dirty state of pages in the kernel buffer. For
+// dirty pages the write() method will be called regardless of having been retrieved
+// previously.
+type NotifyRetrieve struct {
+	Inode  InodeID
+	Unique uint64
+	Offset uint64
+	Length uint32
+}
+
+// Matches the size of WriteIn
+type NotifyRetrieveReplyOp struct {
+	Inode  InodeID
+	Unique uint64
+	Offset uint64
+	Length uint32
+
+	OpContext OpContext
+}

fuseops/simple_types.go

@@ -27,7 +27,7 @@ import (
 // RootInodeID.
 //
 // This corresponds to struct inode::i_no in the VFS layer.
-// (Cf. http://goo.gl/tvYyQt)
+// (https://tinyurl.com/23sr9svd)
 type InodeID uint64
 
 // RootInodeID is a distinguished inode ID that identifies the root of the file
@@ -57,7 +57,7 @@ func init() {
 }
 
 // InodeAttributes contains attributes for a file or directory inode. It
-// corresponds to struct inode (cf. http://goo.gl/tvYyQt).
+// corresponds to struct inode (https://tinyurl.com/23sr9svd).
 type InodeAttributes struct {
 	Size uint64
 
@@ -70,23 +70,26 @@ type InodeAttributes struct {
 	// Note that in contrast to the defaults for FUSE, this package mounts file
 	// systems in a manner such that the kernel checks inode permissions in the
 	// standard posix way. This is implemented by setting the default_permissions
-	// mount option (cf. http://goo.gl/1LxOop and http://goo.gl/1pTjuk).
+	// mount option (https://tinyurl.com/ytun2zsn, https://tinyurl.com/52hz9vya).
 	//
 	// For example, in the case of mkdir:
 	//
-	//  *  (http://goo.gl/JkdxDI) sys_mkdirat calls inode_permission.
+	//  *  (https://tinyurl.com/4yp9bu3h) sys_mkdirat calls inode_permission.
 	//
 	//  *  (...) inode_permission eventually calls do_inode_permission.
 	//
-	//  *  (http://goo.gl/aGCsmZ) calls i_op->permission, which is
-	//     fuse_permission (cf. http://goo.gl/VZ9beH).
+	//  *  (https://tinyurl.com/5f9k2eya) calls i_op->permission, which is
+	//     fuse_permission (https://tinyurl.com/4kevbw27).
 	//
-	//  *  (http://goo.gl/5kqUKO) fuse_permission doesn't do anything at all for
-	//     several code paths if FUSE_DEFAULT_PERMISSIONS is unset. In contrast,
-	//     if that flag *is* set, then it calls generic_permission.
+	//  *  (https://tinyurl.com/nfea3pwj) fuse_permission doesn't do anything at
+	//     all for several code paths if FUSE_DEFAULT_PERMISSIONS is unset. In
+	//     contrast, if that flag *is* set, then it calls generic_permission.
 	//
 	Mode os.FileMode
 
+	// The device number. Only valid if the file is a device
+	Rdev uint32
+
 	// Time information. See `man 2 stat` for full details.
 	Atime  time.Time // Time of last access
 	Mtime  time.Time // Time of last modification
@@ -114,15 +117,15 @@ func (a *InodeAttributes) DebugString() string {
 // when an ID is reused.
 //
 // This corresponds to struct inode::i_generation in the VFS layer.
-// (Cf. http://goo.gl/tvYyQt)
+// (https://tinyurl.com/23sr9svd)
 //
 // Some related reading:
 //
 //	http://fuse.sourceforge.net/doxygen/structfuse__entry__param.html
 //	http://stackoverflow.com/q/11071996/1505451
-//	http://goo.gl/CqvwyX
+//	https://tinyurl.com/yn7wmcmy
 //	http://julipedia.meroh.net/2005/09/nfs-file-handles.html
-//	http://goo.gl/wvo3MB
+//	https://tinyurl.com/2c8vsfrs
 type GenerationNumber uint64
 
 // HandleID is an opaque 64-bit number used to identify a particular open
@@ -157,7 +160,7 @@ type ChildInodeEntry struct {
 	// Ownership information in particular must be set to something reasonable or
 	// by default root will own everything and unprivileged users won't be able
 	// to do anything useful. In traditional file systems in the kernel, the
-	// function inode_init_owner (http://goo.gl/5qavg8) contains the
+	// function inode_init_owner (https://tinyurl.com/5yfdrfdf) contains the
 	// standards-compliant logic for this.
 	Attributes InodeAttributes
 
@@ -166,16 +169,19 @@ type ChildInodeEntry struct {
 	//
 	// For example, this is the abridged call chain for fstat(2):
 	//
-	//  *  (http://goo.gl/tKBH1p) fstat calls vfs_fstat.
-	//  *  (http://goo.gl/3HeITq) vfs_fstat eventuall calls vfs_getattr_nosec.
-	//  *  (http://goo.gl/DccFQr) vfs_getattr_nosec calls i_op->getattr.
-	//  *  (http://goo.gl/dpKkst) fuse_getattr calls fuse_update_attributes.
-	//  *  (http://goo.gl/yNlqPw) fuse_update_attributes uses the values in the
-	//     struct inode if allowed, otherwise calling out to the user-space code.
+	//  *  (https://tinyurl.com/bdd6ek3c) fstat calls vfs_fstat.
+	//  *  (https://tinyurl.com/3enne935) vfs_fstat eventuall calls
+	//     vfs_getattr_nosec.
+	//  *  (https://tinyurl.com/y5rkhzx4) vfs_getattr_nosec calls i_op->getattr.
+	//  *  (https://tinyurl.com/33hawubc) fuse_getattr calls
+	//     fuse_update_attributes.
+	//  *  (https://tinyurl.com/ywhhshxt) fuse_update_attributes uses the values
+	//      in the struct inode if allowed, otherwise calling out to the
+	//      user-space code.
 	//
 	// In addition to obvious cases like fstat, this is also used in more subtle
-	// cases like updating size information before seeking (http://goo.gl/2nnMFa)
-	// or reading (http://goo.gl/FQSWs8).
+	// cases like updating size information before seeking
+	// (https://tinyurl.com/hv2jabnh) or reading (https://tinyurl.com/bdkpz96v).
 	//
 	// Most 'real' file systems do not set inode_operations::getattr, and
 	// therefore vfs_getattr_nosec calls generic_fillattr which simply grabs the
@@ -202,16 +208,18 @@ type ChildInodeEntry struct {
 	// As in the discussion of attribute caching above, unlike real file systems,
 	// FUSE file systems may spontaneously change their name -> inode mapping.
 	// Therefore the FUSE VFS layer uses dentry_operations::d_revalidate
-	// (http://goo.gl/dVea0h) to intercept lookups and revalidate by calling the
-	// user-space LookUpInode method. However the latter may be slow, so it
-	// caches the entries until the time defined by this field.
+	// (https://tinyurl.com/ydb8ncrk) to intercept lookups and revalidate by
+	// calling the user-space LookUpInode method. However the latter may be slow,
+	// so it caches the entries until the time defined by this field.
 	//
 	// Example code walk:
 	//
-	//     * (http://goo.gl/M2G3tO) lookup_dcache calls d_revalidate if enabled.
-	//     * (http://goo.gl/ef0Elu) fuse_dentry_revalidate just uses the dentry's
-	//     inode if fuse_dentry_time(entry) hasn't passed. Otherwise it sends a
-	//     lookup request.
+	//     * (https://tinyurl.com/crddueft) lookup_dcache calls d_revalidate if
+	//       enabled.
+	//
+	//     * (https://tinyurl.com/bdsxacjy) fuse_dentry_revalidate just uses the
+	//       dentry's inode if fuse_dentry_time(entry) hasn't passed. Otherwise
+	//       it sends a lookup request.
 	//
 	// Leave at the zero value to disable caching.
 	//

fusetesting/readdir.go

@@ -30,7 +30,7 @@ func (f sortedEntries) Swap(i, j int)      { f[i], f[j] = f[j], f[i] }
 // Read the directory with the given name and return a list of directory
 // entries, sorted by name.
 //
-// Unlike ioutil.ReadDir (cf. http://goo.gl/i0nNP4), this function does not
+// Unlike ioutil.ReadDir (https://tinyurl.com/yft8kkxb), this function does not
 // silently ignore "file not found" errors when stat'ing the names read from
 // the directory.
 func ReadDirPicky(dirname string) (entries []os.FileInfo, err error) {

fusetesting/stat_linux.go

@@ -28,7 +28,7 @@ func extractBirthtime(sys interface{}) (birthtime time.Time, ok bool) {
 }
 
 func extractNlink(sys interface{}) (nlink uint64, ok bool) {
-	return sys.(*syscall.Stat_t).Nlink, true
+	return uint64(sys.(*syscall.Stat_t).Nlink), true
 }
 
 func getTimes(stat *syscall.Stat_t) (atime, ctime, mtime time.Time) {

fuseutil/dirent.go

@@ -19,6 +19,7 @@ import (
 	"unsafe"
 
 	"github.com/jacobsa/fuse/fuseops"
+	"github.com/jacobsa/fuse/internal/fusekernel"
 )
 
 type DirentType uint32
@@ -50,14 +51,22 @@ type Dirent struct {
 	Type DirentType
 }
 
-// Write the supplied directory entry intto the given buffer in the format
-// expected in fuseops.ReadFileOp.Data, returning the number of bytes written.
+// Write the supplied directory entry into the given buffer in the format
+// expected in fuseops.ReadDirOp.Data, returning the number of bytes written.
 // Return zero if the entry would not fit.
 func WriteDirent(buf []byte, d Dirent) (n int) {
+	return WriteDirentPlus(buf, nil, d)
+}
+
+// Write the supplied directory entry and, optionally, inode entry into the
+// given buffer in the format expected in fuseops.ReadDirOp.Data with enabled
+// READDIRPLUS capability, returning the number of bytes written.
+// Returns zero if the entry would not fit.
+func WriteDirentPlus(buf []byte, e *fuseops.ChildInodeEntry, d Dirent) (n int) {
 	// We want to write bytes with the layout of fuse_dirent
-	// (http://goo.gl/BmFxob) in host order. The struct must be aligned according
-	// to FUSE_DIRENT_ALIGN (http://goo.gl/UziWvH), which dictates 8-byte
-	// alignment.
+	// (https://tinyurl.com/4k7y2h9r) in host order. The struct must be aligned
+	// according to FUSE_DIRENT_ALIGN (https://tinyurl.com/3m3ewu7h), which
+	// dictates 8-byte alignment.
 	type fuse_dirent struct {
 		ino     uint64
 		off     uint64
@@ -78,10 +87,21 @@ func WriteDirent(buf []byte, d Dirent) (n int) {
 
 	// Do we have enough room?
 	totalLen := direntSize + len(d.Name) + padLen
+	if e != nil {
+		// READDIRPLUS was added in protocol 7.21, entry attributes were added in 7.9
+		// So here EntryOut is always full-length
+		totalLen += int(unsafe.Sizeof(fusekernel.EntryOut{}))
+	}
 	if totalLen > len(buf) {
 		return n
 	}
 
+	if e != nil {
+		out := (*fusekernel.EntryOut)(unsafe.Pointer(&buf[n]))
+		fuseops.ConvertChildInodeEntry(e, out)
+		n += int(unsafe.Sizeof(fusekernel.EntryOut{}))
+	}
+
 	// Write the header.
 	de := fuse_dirent{
 		ino:     uint64(d.Inode),

fuseutil/file_system.go

@@ -63,6 +63,10 @@ type FileSystem interface {
 	ListXattr(context.Context, *fuseops.ListXattrOp) error
 	SetXattr(context.Context, *fuseops.SetXattrOp) error
 	Fallocate(context.Context, *fuseops.FallocateOp) error
+	SyncFS(context.Context, *fuseops.SyncFSOp) error
+	Poll(context.Context, *fuseops.PollOp) error
+
+	SetConnection(*fuse.Connection)
 
 	// Regard all inodes (including the root inode) as having their lookup counts
 	// decremented to zero, and clean up any resources associated with the file
@@ -81,8 +85,8 @@ type FileSystem interface {
 //
 // (It is safe to naively process ops concurrently because the kernel
 // guarantees to serialize operations that the user expects to happen in order,
-// cf. http://goo.gl/jnkHPO, fuse-devel thread "Fuse guarantees on concurrent
-// requests").
+// cf. https://tinyurl.com/bddm85v5, fuse-devel thread "Fuse guarantees on
+// concurrent requests").
 func NewFileSystemServer(fs FileSystem) fuse.Server {
 	return &fileSystemServer{
 		fs: fs,
@@ -95,6 +99,8 @@ type fileSystemServer struct {
 }
 
 func (s *fileSystemServer) ServeOps(c *fuse.Connection) {
+	s.fs.SetConnection(c)
+
 	// When we are done, we clean up by waiting for all in-flight ops then
 	// destroying the file system.
 	defer func() {
@@ -236,6 +242,12 @@ func (s *fileSystemServer) handleOp(
 
 	case *fuseops.FallocateOp:
 		err = s.fs.Fallocate(ctx, typed)
+
+	case *fuseops.SyncFSOp:
+		err = s.fs.SyncFS(ctx, typed)
+
+	case *fuseops.PollOp:
+		err = s.fs.Poll(ctx, typed)
 	}
 
 	c.Reply(ctx, err)

fuseutil/not_implemented_file_system.go

@@ -204,5 +204,20 @@ func (fs *NotImplementedFileSystem) Fallocate(
 	return fuse.ENOSYS
 }
 
+func (fs *NotImplementedFileSystem) SyncFS(
+	ctx context.Context,
+	op *fuseops.SyncFSOp) error {
+	return fuse.ENOSYS
+}
+
+func (fs *NotImplementedFileSystem) Poll(
+	ctx context.Context,
+	op *fuseops.PollOp) error {
+	return fuse.ENOSYS
+}
+
+func (fs *NotImplementedFileSystem) SetConnection(*fuse.Connection) {
+}
+
 func (fs *NotImplementedFileSystem) Destroy() {
 }

go.mod

@@ -1,17 +1,19 @@
 module github.com/jacobsa/fuse
 
-go 1.18
+go 1.20
 
 require (
 	github.com/detailyang/go-fallocate v0.0.0-20180908115635-432fa640bd2e
 	github.com/jacobsa/oglematchers v0.0.0-20150720000706-141901ea67cd
-	github.com/jacobsa/oglemock v0.0.0-20150831005832-e94d794d06ff // indirect
 	github.com/jacobsa/ogletest v0.0.0-20170503003838-80d50a735a11
-	github.com/jacobsa/reqtrace v0.0.0-20150505043853-245c9e0234cb // indirect
 	github.com/jacobsa/syncutil v0.0.0-20180201203307-228ac8e5a6c3
 	github.com/jacobsa/timeutil v0.0.0-20170205232429-577e5acbbcf6
 	github.com/kylelemons/godebug v1.1.0
-	golang.org/x/crypto v0.0.0-20220525230936-793ad666bf5e // indirect
-	golang.org/x/net v0.0.0-20220526153639-5463443f8c37
-	golang.org/x/sys v0.0.0-20220520151302-bc2c85ada10a
+	golang.org/x/net v0.23.0
+	golang.org/x/sys v0.18.0
+)
+
+require (
+	github.com/jacobsa/oglemock v0.0.0-20150831005832-e94d794d06ff // indirect
+	github.com/jacobsa/reqtrace v0.0.0-20150505043853-245c9e0234cb // indirect
 )

go.sum

@@ -14,25 +14,7 @@ github.com/jacobsa/timeutil v0.0.0-20170205232429-577e5acbbcf6 h1:XKHJmHcgU9glxk
 github.com/jacobsa/timeutil v0.0.0-20170205232429-577e5acbbcf6/go.mod h1:JEWKD6V8xETMW+DEv+IQVz++f8Cn8O/X0HPeDY3qNis=
 github.com/kylelemons/godebug v1.1.0 h1:RPNrshWIDI6G2gRW9EHilWtl7Z6Sb1BR0xunSBf0SNc=
 github.com/kylelemons/godebug v1.1.0/go.mod h1:9/0rRGxNHcop5bhtWyNeEfOS8JIWk580+fNqagV/RAw=
-golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
-golang.org/x/crypto v0.0.0-20220525230936-793ad666bf5e h1:T8NU3HyQ8ClP4SEE+KbFlg6n0NhuTsN4MyznaarGsZM=
-golang.org/x/crypto v0.0.0-20220525230936-793ad666bf5e/go.mod h1:IxCIyHEi3zRg3s0A5j5BB6A9Jmi73HwBIUl50j+osU4=
-golang.org/x/net v0.0.0-20200301022130-244492dfa37a h1:GuSPYbZzB5/dcLNCwLQLsg3obCJtX9IJhpXkvY7kzk0=
-golang.org/x/net v0.0.0-20200301022130-244492dfa37a/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
-golang.org/x/net v0.0.0-20211112202133-69e39bad7dc2/go.mod h1:9nx3DQGgdP8bBQD5qxJ1jj9UTztislL4KSBs9R2vV5Y=
-golang.org/x/net v0.0.0-20220526153639-5463443f8c37 h1:lUkvobShwKsOesNfWWlCS5q7fnbG1MEliIzwu886fn8=
-golang.org/x/net v0.0.0-20220526153639-5463443f8c37/go.mod h1:XRhObCWvk6IyKnWLug+ECip1KBveYUHfp+8e9klMJ9c=
-golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
-golang.org/x/sys v0.0.0-20200302150141-5c8b2ff67527 h1:uYVVQ9WP/Ds2ROhcaGPeIdVq0RIXVLwsHlnvJ+cT1So=
-golang.org/x/sys v0.0.0-20200302150141-5c8b2ff67527/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20210423082822-04245dca01da/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20210615035016-665e8c7367d1/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.0.0-20220520151302-bc2c85ada10a h1:dGzPydgVsqGcTRVwiLJ1jVbufYwmzD3LfVPLKsKg+0k=
-golang.org/x/sys v0.0.0-20220520151302-bc2c85ada10a/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
-golang.org/x/term v0.0.0-20210927222741-03fcf44c2211/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
-golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
-golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/text v0.3.7/go.mod h1:u+2+/6zg+i71rQMx5EYifcz6MCKuco9NR6JIITiCfzQ=
-golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/net v0.23.0 h1:7EYJ93RZ9vYSZAIb2x3lnuvqO5zneoD6IvWjuhfxjTs=
+golang.org/x/net v0.23.0/go.mod h1:JKghWKKOSdJwpW2GEx0Ja7fmaKnMsbu+MWVZTokSYmg=
+golang.org/x/sys v0.18.0 h1:DBdB3niSjOA/O0blCZBqDefyWNYveAYMNF1Wum0DYQ4=
+golang.org/x/sys v0.18.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=

internal/buffer/in_message.go

@@ -17,6 +17,7 @@ package buffer
 import (
 	"fmt"
 	"io"
+	"sync"
 	"syscall"
 	"unsafe"
 
@@ -52,11 +53,38 @@ func NewInMessage() *InMessage {
 	}
 }
 
+var readLock sync.Mutex
+
+func (m *InMessage) ReadSingle(r io.Reader) (int, error) {
+	readLock.Lock()
+	defer readLock.Unlock()
+
+	// read request length
+	if _, err := io.ReadFull(r, m.storage[0:4]); err != nil {
+		return 0, err
+	}
+
+	l := m.Header().Len
+	// read remaining request
+	if n, err := io.ReadFull(r, m.storage[4:l]); err != nil {
+		return n, err
+	}
+	return int(l), nil
+}
+
 // Initialize with the data read by a single call to r.Read. The first call to
 // Consume will consume the bytes directly after the fusekernel.InHeader
 // struct.
 func (m *InMessage) Init(r io.Reader) error {
-	n, err := r.Read(m.storage[:])
+
+	var n int
+	var err error
+	if fusekernel.IsPlatformFuseT {
+		n, err = m.ReadSingle(r)
+	} else {
+		n, err = r.Read(m.storage[:])
+	}
+
 	if err != nil {
 		return err
 	}

internal/buffer/in_message_linux.go

@@ -17,4 +17,4 @@ package buffer
 // The maximum fuse write request size that InMessage can acommodate.
 //
 // As of kernel 4.20 Linux accepts writes up to 256 pages or 1MiB
-const MaxWriteSize = 1 << 20
+const MaxWriteSize = 1 << 17

internal/fusekernel/fuse_kernel.go

@@ -41,12 +41,14 @@ import (
 	"unsafe"
 )
 
+var IsPlatformFuseT bool
+
 // The FUSE version implemented by the package.
 const (
 	ProtoVersionMinMajor = 7
-	ProtoVersionMinMinor = 19
+	ProtoVersionMinMinor = 18
 	ProtoVersionMaxMajor = 7
-	ProtoVersionMaxMinor = 31
+	ProtoVersionMaxMinor = 34
 )
 
 const (
@@ -168,7 +170,7 @@ const (
 
 // OpenAccessModeMask is a bitmask that separates the access mode
 // from the other flags in OpenFlags.
-const OpenAccessModeMask OpenFlags = syscall.O_ACCMODE
+const OpenAccessModeMask OpenFlags = OpenReadOnly | OpenWriteOnly | OpenReadWrite
 
 // OpenFlags are the O_FOO flags passed to open/create/etc calls. For
 // example, os.O_WRONLY | os.O_APPEND.
@@ -227,6 +229,7 @@ const (
 	OpenDirectIO    OpenResponseFlags = 1 << 0 // bypass page cache for this open file
 	OpenKeepCache   OpenResponseFlags = 1 << 1 // don't invalidate the data cache on open
 	OpenNonSeekable OpenResponseFlags = 1 << 2 // mark the file as non-seekable (not supported on OS X)
+	OpenCacheDir    OpenResponseFlags = 1 << 3 // allow caching this directory
 
 	OpenPurgeAttr OpenResponseFlags = 1 << 30 // OS X
 	OpenPurgeUBC  OpenResponseFlags = 1 << 31 // OS X
@@ -240,6 +243,7 @@ var openResponseFlagNames = []flagName{
 	{uint32(OpenDirectIO), "OpenDirectIO"},
 	{uint32(OpenKeepCache), "OpenKeepCache"},
 	{uint32(OpenNonSeekable), "OpenNonSeekable"},
+	{uint32(OpenCacheDir), "OpenCacheDir"},
 	{uint32(OpenPurgeAttr), "OpenPurgeAttr"},
 	{uint32(OpenPurgeUBC), "OpenPurgeUBC"},
 }
@@ -266,6 +270,7 @@ const (
 	InitAsyncDIO         InitFlags = 1 << 15
 	InitWritebackCache   InitFlags = 1 << 16
 	InitNoOpenSupport    InitFlags = 1 << 17
+	InitParallelDirOps   InitFlags = 1 << 18
 	InitMaxPages         InitFlags = 1 << 22
 	InitCacheSymlinks    InitFlags = 1 << 23
 	InitNoOpendirSupport InitFlags = 1 << 24
@@ -346,6 +351,34 @@ var releaseFlagNames = []flagName{
 	{uint32(ReleaseFlush), "ReleaseFlush"},
 }
 
+// Poll flags and events are used in the Poll exchange.
+type PollFlags uint32
+
+const (
+	// From the kernel source:
+	// Ask for notification if there's someone waiting for it.
+	// The client may ignore the flag and always notify.
+	PollScheduleNotify PollFlags = 1 << 0
+)
+
+type PollEvents uint32
+
+const (
+	PollInEvent     PollEvents = 0x0001
+	PollPriEvent    PollEvents = 0x0002
+	PollOutEvent    PollEvents = 0x0004
+	PollErrEvent    PollEvents = 0x0008
+	PollHupEvent    PollEvents = 0x0010
+	PollNvalEvent   PollEvents = 0x0020
+	PollRdNormEvent PollEvents = 0x0040
+	PollRdBandEvent PollEvents = 0x0080
+	PollWrNormEvent PollEvents = 0x0100
+	PollWrBandEvent PollEvents = 0x0200
+	PollMsgEvent    PollEvents = 0x0400
+	PollRemoveEvent PollEvents = 0x1000
+	PollRdHupEvent  PollEvents = 0x2000
+)
+
 // Opcodes
 const (
 	OpLookup      = 1
@@ -386,8 +419,17 @@ const (
 	OpDestroy     = 38
 	OpIoctl       = 39 // Linux?
 	OpPoll        = 40 // Linux?
+	OpNotifyReply = 41
 	OpBatchForget = 42
 	OpFallocate   = 43
+	OpReaddirplus = 44
+	//
+	OpRename2       = 45
+	OpLseek         = 46
+	OpCopyFileRange = 47
+	OpSetupMapping  = 48
+	OpRemoveMapping = 49
+	OpSyncFS        = 50
 
 	// OS X
 	OpSetvolname = 61
@@ -552,6 +594,18 @@ func CreateInSize(p Protocol) uintptr {
 	}
 }
 
+type PollIn struct {
+	Fh     uint64
+	Kh     uint64
+	Flags  uint32
+	Events uint32
+}
+
+type PollOut struct {
+	Revents uint32
+	padding uint32
+}
+
 type ReleaseIn struct {
 	Fh           uint64
 	Flags        uint32
@@ -787,8 +841,15 @@ const (
 	NotifyCodePoll       int32 = 1
 	NotifyCodeInvalInode int32 = 2
 	NotifyCodeInvalEntry int32 = 3
+	NotifyCodeStore      int32 = 4
+	NotifyCodeRetrieve   int32 = 5
+	NotifyCodeDelete     int32 = 6
 )
 
+type NotifyPollWakeupOut struct {
+	Kh uint64
+}
+
 type NotifyInvalInodeOut struct {
 	Ino uint64
 	Off int64
@@ -800,3 +861,39 @@ type NotifyInvalEntryOut struct {
 	Namelen uint32
 	padding uint32
 }
+
+type SyncFSIn struct {
+	Padding uint64
+}
+
+type NotifyDeleteOut struct {
+	Parent  uint64
+	Child   uint64
+	Namelen uint32
+	padding uint32
+}
+
+type NotifyStoreOut struct {
+	Nodeid  uint64
+	Offset  uint64
+	Size    uint32
+	padding uint32
+}
+
+type NotifyRetrieveOut struct {
+	Unique  uint64
+	Nodeid  uint64
+	Offset  uint64
+	Size    uint32
+	padding uint32
+}
+
+// Matches the size of WriteIn
+type NotifyRetrieveIn struct {
+	dummy1 uint64
+	Offset uint64
+	Size   uint32
+	dummy2 uint32
+	dummy3 uint64
+	dummy4 uint64
+}

internal/fusekernel/fuse_kernel_default.go

@@ -1,3 +1,5 @@
+// +build linux windows
+
 package fusekernel
 
 import "time"

internal/fusekernel/fuse_kernel_std.go

@@ -1 +0,0 @@
-package fusekernel

mount_config.go

@@ -50,7 +50,7 @@ type MountConfig struct {
 	// Linux only. OS X always behaves as if writeback caching is disabled.
 	//
 	// By default on Linux we allow the kernel to perform writeback caching
-	// (cf. http://goo.gl/LdZzo1):
+	// (https://tinyurl.com/3ma8ypeu):
 	//
 	// *   When the user calls write(2), the kernel sticks the user's data into
 	//     its page cache. Only later does it call through to the file system,
@@ -65,7 +65,7 @@ type MountConfig struct {
 	//
 	// *   close(2) (and anything else calling f_op->flush) causes all dirty
 	//     pages to be written out before it proceeds to send a FlushFileOp
-	//     (cf. https://goo.gl/TMrY6X).
+	//     (https://tinyurl.com/3ur6vmsv).
 	//
 	// *   Similarly, close(2) causes the kernel to send a setattr request
 	//     filling in the mtime if any dirty pages were flushed, since the time
@@ -78,22 +78,23 @@ type MountConfig struct {
 	//
 	//     Code walk:
 	//
-	//     *   (https://goo.gl/zTIZQ9) fuse_flush calls write_inode_now before
-	//         calling the file system. The latter eventually calls into
+	//     *   (https://tinyurl.com/3ur6vmsv) fuse_flush calls write_inode_now
+	//         before calling the file system. The latter eventually calls into
 	//         __writeback_single_inode.
 	//
-	//     *   (https://goo.gl/L7Z2w5) __writeback_single_inode calls
+	//     *   (https://tinyurl.com/35vtmtsz) __writeback_single_inode calls
 	//         do_writepages, which writes out any dirty pages.
 	//
-	//     *   (https://goo.gl/DOPgla) __writeback_single_inode later calls
-	//         write_inode, which calls into the superblock op struct's write_inode
-	//         member. For fuse, this is fuse_write_inode
-	//         (cf. https://goo.gl/eDSKOX).
+	//     *   (https://tinyurl.com/3wv4paaf) __writeback_single_inode later
+	//         calls write_inode, which calls into the superblock op struct's
+	//         write_inode member. For fuse, this is fuse_write_inode
+	//         (https://tinyurl.com/mrxupe98).
 	//
-	//     *   (https://goo.gl/PbkGA1) fuse_write_inode calls fuse_flush_times.
+	//     *   (https://tinyurl.com/mrxt9bta) fuse_write_inode calls
+	//         fuse_flush_times.
 	//
-	//     *   (https://goo.gl/ig8x9V) fuse_flush_times sends a setttr request
-	//         for setting the inode's mtime.
+	//     *   (https://tinyurl.com/mr49cjdf) fuse_flush_times sends a setttr
+	//         request for setting the inode's mtime.
 	//
 	// However, this brings along some caveats:
 	//
@@ -102,11 +103,11 @@ type MountConfig struct {
 	//
 	// *   The kernel caches mtime and ctime regardless of whether the file
 	//     system tells it to do so, disregarding the result of further getattr
-	//     requests (cf. https://goo.gl/3ZZMUw, https://goo.gl/7WtQUp). It
-	//     appears this may be true of the file size, too. Writeback caching may
-	//     therefore not be suitable for file systems where these attributes can
-	//     spontaneously change for reasons the kernel doesn't observe. See
-	//     http://goo.gl/V5WQCN for more discussion.
+	//     requests (https://tinyurl.com/mrxnfatv, https://tinyurl.com/27jju8n4).
+	//     It appears this may be true of the file size, too. Writeback caching
+	//     may therefore not be suitable for file systems where these attributes
+	//     can spontaneously change for reasons the kernel doesn't observe. See
+	//     https://tinyurl.com/yyprvjvs for more discussion.
 	//
 	// Setting DisableWritebackCaching disables this behavior. Instead the file
 	// system is called one or more times for each write(2), and the user's
@@ -116,12 +117,12 @@ type MountConfig struct {
 	// OS X only.
 	//
 	// Normally on OS X we mount with the novncache option
-	// (cf. http://goo.gl/1pTjuk), which disables entry caching in the kernel.
-	// This is because osxfuse does not honor the entry expiration values we
-	// return to it, instead caching potentially forever (cf.
-	// http://goo.gl/8yR0Ie), and it is probably better to fail to cache than to
-	// cache for too long, since the latter is more likely to hide consistency
-	// bugs that are difficult to detect and diagnose.
+	// (https://tinyurl.com/52hz9vya), which disables entry caching in the
+	// kernel. This is because macFUSE (osxfuse) does not honor the entry
+	// expiration values we return to it, instead caching potentially forever
+	// (https://tinyurl.com/2rr6cd3m), and it is probably better to fail to cache
+	// than to cache for too long, since the latter is more likely to hide
+	// consistency bugs that are difficult to detect and diagnose.
 	//
 	// This field disables the use of novncache, restoring entry caching. Beware:
 	// the value of ChildInodeEntry.EntryExpiration is ignored by the kernel, and
@@ -151,6 +152,11 @@ type MountConfig struct {
 	// OpenDir calls at all (Linux >= 5.1):
 	EnableNoOpendirSupport bool
 
+	// Tell the kernel to use READDIRPLUS.
+	// Note that the implementation may still fall back to READDIR if the running
+	// kernel doesn't have support for READDIRPLUS.
+	UseReadDirPlus bool
+
 	// Disable FUSE default permissions.
 	// This is useful for situations where the backing data store (e.g., S3) doesn't
 	// actually utilise any form of qualifiable UNIX permissions.
@@ -166,9 +172,16 @@ type MountConfig struct {
 	// OS X only.
 	//
 	// The name of the mounted volume, as displayed in the Finder. If empty, a
-	// default name involving the string 'osxfuse' is used.
+	// default name involving the string 'osxfuse' (the old name of macFUSE)
+	// is used.
 	VolumeName string
 
+	// OS X only.
+	//
+	// The FUSE implementation to use. One of FUSEImplFuseT (default) or
+	// FUSEImplMacFUSE.
+	FuseImpl FUSEImpl
+
 	// Additional key=value options to pass unadulterated to the underlying mount
 	// command. See `man 8 mount`, the fuse documentation, etc. for
 	// system-specific information.
@@ -185,8 +198,20 @@ type MountConfig struct {
 	// Flag to enable async reads that are received from
 	// the kernel
 	EnableAsyncReads bool
+
+	// Flag to enable parallel lookup and readdir operations from the
+	// kernel
+	// Ref: https://github.com/torvalds/linux/commit/5c672ab3f0ee0f78f7acad183f34db0f8781a200
+	EnableParallelDirOps bool
 }
 
+type FUSEImpl uint8
+
+const (
+	FUSEImplFuseT = iota
+	FUSEImplMacFUSE
+)
+
 // Create a map containing all of the key=value mount options to be given to
 // the mount helper.
 func (c *MountConfig) toMap() (opts map[string]string) {

mount_darwin.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"errors"
 	"fmt"
+	"log"
 	"os"
 	"os/exec"
 	"strconv"
@@ -11,6 +12,7 @@ import (
 	"syscall"
 
 	"github.com/jacobsa/fuse/internal/buffer"
+	"github.com/jacobsa/fuse/internal/fusekernel"
 )
 
 var errNoAvail = errors.New("no available fuse devices")
@@ -78,6 +80,8 @@ var (
 	}
 )
 
+const FUSET_SRV_PATH = "/usr/local/bin/go-nfsv4"
+
 func loadOSXFUSE(bin string) error {
 	cmd := exec.Command(bin)
 	cmd.Dir = "/"
@@ -212,7 +216,7 @@ func callMountCommFD(
 // to the kernel. Mounting continues in the background, and is complete when an
 // error is written to the supplied channel. The file system may need to
 // service the connection in order for mounting to complete.
-func mount(
+func mountOsxFuse(
 	dir string,
 	cfg *MountConfig,
 	ready chan<- error) (dev *os.File, err error) {
@@ -263,3 +267,161 @@ func mount(
 
 	return nil, errOSXFUSENotFound
 }
+
+func fusetBinary() (string, error) {
+	srv_path := os.Getenv("FUSE_NFSSRV_PATH")
+	if srv_path == "" {
+		srv_path = FUSET_SRV_PATH
+	}
+
+	if _, err := os.Stat(srv_path); err == nil {
+		return srv_path, nil
+	}
+
+	return "", fmt.Errorf("FUSE-T not found")
+}
+
+func unixgramSocketpair() (l, r *os.File, err error) {
+	fd, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0)
+	if err != nil {
+		return nil, nil, os.NewSyscallError("socketpair",
+			err.(syscall.Errno))
+	}
+	l = os.NewFile(uintptr(fd[0]), fmt.Sprintf("socketpair-half%d", fd[0]))
+	r = os.NewFile(uintptr(fd[1]), fmt.Sprintf("socketpair-half%d", fd[1]))
+	return
+}
+
+var local, local_mon, remote, remote_mon *os.File
+
+func startFuseTServer(binary string, argv []string,
+	additionalEnv []string,
+	wait bool,
+	debugLogger *log.Logger,
+	ready chan<- error) (*os.File, error) {
+	if debugLogger != nil {
+		debugLogger.Println("Creating a socket pair")
+	}
+
+	var err error
+	local, remote, err = unixgramSocketpair()
+	if err != nil {
+		return nil, err
+	}
+	defer remote.Close()
+
+	local_mon, remote_mon, err = unixgramSocketpair()
+	if err != nil {
+		return nil, err
+	}
+	defer remote_mon.Close()
+
+	syscall.CloseOnExec(int(local.Fd()))
+	syscall.CloseOnExec(int(local_mon.Fd()))
+
+	if debugLogger != nil {
+		debugLogger.Println("Creating files to wrap the sockets")
+	}
+
+	if debugLogger != nil {
+		debugLogger.Println("Starting fusermount/os mount")
+	}
+	// Start fusermount/mount_macfuse/mount_osxfuse.
+	cmd := exec.Command(binary, argv...)
+	cmd.Env = append(os.Environ(), "_FUSE_COMMFD=3")
+	cmd.Env = append(cmd.Env, "_FUSE_MONFD=4")
+	cmd.Env = append(cmd.Env, additionalEnv...)
+	cmd.ExtraFiles = []*os.File{remote, remote_mon}
+	cmd.Stderr = nil
+	cmd.Stdout = nil
+	// daemonize
+	cmd.SysProcAttr = &syscall.SysProcAttr{
+		Setsid: true,
+	}
+
+	// Run the command.
+	err = cmd.Start()
+	cmd.Process.Release()
+	if err != nil {
+		return nil, fmt.Errorf("running %v: %v", binary, err)
+	}
+
+	if debugLogger != nil {
+		debugLogger.Println("Wrapping socket pair in a connection")
+	}
+
+	if debugLogger != nil {
+		debugLogger.Println("Checking that we have a unix domain socket")
+	}
+
+	if debugLogger != nil {
+		debugLogger.Println("Read a message from socket")
+	}
+
+	go func() {
+		if _, err = local_mon.Write([]byte("mount")); err != nil {
+			err = fmt.Errorf("fuse-t failed: %v", err)
+		} else {
+			reply := make([]byte, 4)
+			if _, err = local_mon.Read(reply); err != nil {
+				fmt.Printf("mount read  %v\n", err)
+				err = fmt.Errorf("fuse-t failed: %v", err)
+			}
+		}
+
+		ready <- err
+		close(ready)
+	}()
+
+	if debugLogger != nil {
+		debugLogger.Println("Successfully read the socket message.")
+	}
+
+	return local, nil
+}
+
+func mountFuset(
+	dir string,
+	cfg *MountConfig,
+	ready chan<- error) (dev *os.File, err error) {
+	fuseTBin, err := fusetBinary()
+	if err != nil {
+		return nil, err
+	}
+
+	fusekernel.IsPlatformFuseT = true
+	env := []string{}
+	argv := []string{
+		fmt.Sprintf("--rwsize=%d", buffer.MaxWriteSize),
+	}
+
+	if cfg.VolumeName != "" {
+		argv = append(argv, "--volname")
+		argv = append(argv, cfg.VolumeName)
+	}
+	if cfg.ReadOnly {
+		argv = append(argv, "-r")
+	}
+
+	env = append(env, "_FUSE_COMMVERS=2")
+	argv = append(argv, dir)
+
+	return startFuseTServer(fuseTBin, argv, env, false, cfg.DebugLogger, ready)
+}
+
+func mount(
+	dir string,
+	cfg *MountConfig,
+	ready chan<- error) (dev *os.File, err error) {
+
+	fusekernel.IsPlatformFuseT = false
+	switch cfg.FuseImpl {
+	case FUSEImplMacFUSE:
+		dev, err = mountOsxFuse(dir, cfg, ready)
+	case FUSEImplFuseT:
+		fallthrough
+	default:
+		dev, err = mountFuset(dir, cfg, ready)
+	}
+	return
+}

mount_linux.go

@@ -84,6 +84,7 @@ func directmount(dir string, cfg *MountConfig) (*os.File, error) {
 		mountflag = fn(mountflag)
 		delete(opts, k)
 	}
+	fsname := opts["fsname"]
 	delete(opts, "fsname") // handled via fstype mount(2) parameter
 	fstype := "fuse"
 	if subtype, ok := opts["subtype"]; ok {
@@ -96,11 +97,11 @@ func directmount(dir string, cfg *MountConfig) (*os.File, error) {
 		cfg.DebugLogger.Println("Starting the unix mounting")
 	}
 	if err := unix.Mount(
-		cfg.FSName, // source
-		dir,        // target
-		fstype,     // fstype
-		mountflag,  // mountflag
-		data,       // data
+		fsname,    // source
+		dir,       // target
+		fstype,    // fstype
+		mountflag, // mountflag
+		data,      // data
 	); err != nil {
 		if err == syscall.EPERM {
 			return nil, errFallback

samples/cachingfs/caching_fs.go

@@ -48,6 +48,11 @@ const (
 // Each file responds to reads with random contents. SetKeepCache can be used
 // to control whether the response to OpenFileOp tells the kernel to keep the
 // file's data in the page cache or not.
+//
+// Each directory responds to readdir with random entries (different names).
+// SetCacheDir and SetKeepDirCache can be used to control whether the response
+// to OpenDirOp tells the kernel to cache the next response of ReadDirOp
+// in cache, or invalidate the existing cached entry in page cache.
 type CachingFS interface {
 	fuseutil.FileSystem
 
@@ -66,6 +71,14 @@ type CachingFS interface {
 	// Instruct the file system whether or not to reply to OpenFileOp with
 	// FOPEN_KEEP_CACHE set.
 	SetKeepCache(keep bool)
+
+	// Instruct the file system whether or not to reply to OpenDirOp with
+	// FOPEN_KEEP_CACHE set.
+	SetKeepDirCache(keep bool)
+
+	// Instruct the file system whether or not to reply to OpenDirOp with
+	// FOPEN_CACHE_DIR set.
+	SetCacheDir(cacheDir bool)
 }
 
 // Create a file system that issues cacheable responses according to the
@@ -126,6 +139,10 @@ type cachingFS struct {
 	// GUARDED_BY(mu)
 	keepPageCache bool
 
+	// GUARDED_BY(mu)
+	keepDirCache bool
+	cacheDir     bool
+
 	// The current ID of the lowest numbered non-root inode.
 	//
 	// INVARIANT: baseID > fuseops.RootInodeID
@@ -202,6 +219,17 @@ func (fs *cachingFS) barAttrs() fuseops.InodeAttributes {
 	}
 }
 
+func randomString(len int) string {
+	bytes := make([]byte, len)
+	nn, err := rand.Read(bytes)
+
+	if err != nil || len != nn {
+		return ""
+	}
+
+	return string(bytes)
+}
+
 ////////////////////////////////////////////////////////////////////////
 // Public interface
 ////////////////////////////////////////////////////////////////////////
@@ -254,6 +282,22 @@ func (fs *cachingFS) SetKeepCache(keep bool) {
 	fs.keepPageCache = keep
 }
 
+// LOCKS_EXCLUDED(fs.mu)
+func (fs *cachingFS) SetKeepDirCache(keep bool) {
+	fs.mu.Lock()
+	defer fs.mu.Unlock()
+
+	fs.keepDirCache = keep
+}
+
+// LOCKS_EXCLUDED(fs.mu)
+func (fs *cachingFS) SetCacheDir(cacheDir bool) {
+	fs.mu.Lock()
+	defer fs.mu.Unlock()
+
+	fs.cacheDir = cacheDir
+}
+
 ////////////////////////////////////////////////////////////////////////
 // FileSystem methods
 ////////////////////////////////////////////////////////////////////////
@@ -349,6 +393,51 @@ func (fs *cachingFS) GetInodeAttributes(
 func (fs *cachingFS) OpenDir(
 	ctx context.Context,
 	op *fuseops.OpenDirOp) error {
+	fs.mu.Lock()
+	defer fs.mu.Unlock()
+
+	op.CacheDir = fs.cacheDir
+	op.KeepCache = fs.keepDirCache
+
+	return nil
+}
+
+func (fs *cachingFS) ReadDir(
+	ctx context.Context,
+	op *fuseops.ReadDirOp) error {
+
+	entries := []fuseutil.Dirent{
+		{
+			Offset: 1,
+			Inode:  101,
+			Name:   "rdir" + randomString(4),
+			Type:   fuseutil.DT_Directory,
+		},
+		{
+			Offset: 2,
+			Inode:  102,
+			Name:   "rfoo" + randomString(4),
+			Type:   fuseutil.DT_File,
+		},
+	}
+
+	// Grab the range of interest.
+	if op.Offset > fuseops.DirOffset(len(entries)) {
+		return fuse.EIO
+	}
+
+	entries = entries[op.Offset:]
+
+	// Resume at the specified offset into the array.
+	for _, e := range entries {
+		n := fuseutil.WriteDirent(op.Dst[op.BytesRead:], e)
+		if n == 0 {
+			break
+		}
+
+		op.BytesRead += n
+	}
+
 	return nil
 }
 

samples/cachingfs/caching_fs_test.go

@@ -723,3 +723,253 @@ func (t *PageCacheTest) TwoFileHandles_KeepCache() {
 
 	ExpectTrue(bytes.Equal(c1, c3))
 }
+
+////////////////////////////////////////////////////////////////////////
+// Dir cache
+////////////////////////////////////////////////////////////////////////
+
+type DirCacheTest struct {
+	cachingFSTest
+}
+
+var _ SetUpInterface = &DirCacheTest{}
+
+func init() { RegisterTestSuite(&DirCacheTest{}) }
+
+func (t *DirCacheTest) SetUp(ti *TestInfo) {
+	const (
+		lookupEntryTimeout = 0
+		getattrTimeout     = 0
+	)
+
+	t.cachingFSTest.setUp(ti, lookupEntryTimeout, getattrTimeout)
+}
+
+func (t *DirCacheTest) CacheDirAndKeepDirCache() {
+	t.fs.SetCacheDir(true)
+	t.fs.SetKeepDirCache(true)
+
+	// First read, kernel will cache the dir response.
+	f, err := os.Open(path.Join(t.Dir, "dir"))
+	defer f.Close() // Make sure to close specially required in failure scenario.
+	AssertEq(nil, err)
+
+	names1, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names1))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	// Second read, will be served from cache.
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names2, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names2))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertEq(names1[0], names2[0])
+	AssertEq(names1[1], names2[1])
+}
+
+func (t *DirCacheTest) NoCacheDirAndKeepDirCache() {
+	t.fs.SetCacheDir(false)
+	t.fs.SetKeepDirCache(true)
+
+	// First read, no caching since NoCacheDir.
+	f, err := os.Open(path.Join(t.Dir, "dir"))
+	defer f.Close() // Make sure to close specially required in failure scenario.
+	AssertEq(nil, err)
+
+	names1, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names1))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	// Second read, will be served from filesystem, hence different name.
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names2, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names2))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertNe(names1[0], names2[0])
+	AssertNe(names1[1], names2[1])
+}
+
+func (t *DirCacheTest) CacheDirAndNoKeepDirCache() {
+	t.fs.SetCacheDir(true)
+	t.fs.SetKeepDirCache(false)
+
+	// First read, kernel will cache the dir response.
+	f, err := os.Open(path.Join(t.Dir, "dir"))
+	defer f.Close() // Make sure to close specially required in failure scenario.
+	AssertEq(nil, err)
+
+	names1, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names1))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	// Second read, cached response will be invalidated since NoKeepDirCache.
+	// Hence, different names.
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names2, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names2))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertNe(names1[0], names2[0])
+	AssertNe(names1[1], names2[1])
+}
+
+func (t *DirCacheTest) NoCacheDirAndNoKeepDirCache() {
+	t.fs.SetCacheDir(false)
+	t.fs.SetKeepDirCache(false)
+
+	// First read, no caching since NoCacheDir.
+	f, err := os.Open(path.Join(t.Dir, "dir"))
+	defer f.Close() // Make sure to close specially required in failure scenario.
+	AssertEq(nil, err)
+
+	names1, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names1))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	// Second read, will be served from filesystem.
+	// Since NoCacheDir also NoKeepDirCache. Hence, different names.
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names2, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names2))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertNe(names1[0], names2[0])
+	AssertNe(names1[1], names2[1])
+}
+
+func (t *DirCacheTest) CacheDirWithChangingKeepDirCache() {
+	t.fs.SetCacheDir(true)
+	t.fs.SetKeepDirCache(false)
+
+	// First read, kernel will cache the dir response.
+	f, err := os.Open(path.Join(t.Dir, "dir"))
+	defer f.Close() // Make sure to close specially required in failure scenario.
+	AssertEq(nil, err)
+
+	names1, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names1))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	// Cached response will not be invalidated. Hence, served from kernel cache.
+	t.fs.SetKeepDirCache(true)
+	// Second read, will be served from cache
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names2, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names2))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertEq(names1[0], names2[0])
+	AssertEq(names1[1], names2[1])
+
+	// Kernel has cached but invalidated due to NoKeepDirCache.
+	// Hence, third read will be served from filesystem.
+	t.fs.SetKeepDirCache(false)
+
+	// Third read, will be served from filesystem. So, names will be different.
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names3, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names3))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertNe(names2[0], names3[0])
+	AssertNe(names2[1], names3[1])
+}
+
+func (t *DirCacheTest) ChangingCacheDirWithKeepDirCache() {
+	t.fs.SetCacheDir(true)
+	t.fs.SetKeepDirCache(true)
+
+	// First read, kernel will cache the dir response.
+	f, err := os.Open(path.Join(t.Dir, "dir"))
+	defer f.Close() // Make sure to close specially required in failure scenario.
+	AssertEq(nil, err)
+
+	names1, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names1))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	// Cached response will not be invalidated, but also not be served from cache.
+	// Since NoCacheDir so names will be different.
+	t.fs.SetCacheDir(false)
+	// Second read, will be served from filesystem.
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names2, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names2))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertNe(names1[0], names2[0])
+	AssertNe(names1[1], names2[1])
+
+	// Third read will be served from cache.
+	// But first read response is cached, since KeepDirCache.
+	t.fs.SetCacheDir(true)
+
+	// Third read, will be served from filesystem. So, names will be different.
+	f, err = os.Open(path.Join(t.Dir, "dir"))
+	AssertEq(nil, err)
+
+	names3, err := f.Readdirnames(-1)
+	AssertEq(nil, err)
+	AssertEq(2, len(names3))
+
+	err = f.Close()
+	AssertEq(nil, err)
+
+	AssertEq(names1[0], names3[0])
+	AssertEq(names1[1], names3[1])
+}

samples/hellofs/hello_fs.go

@@ -218,7 +218,7 @@ func (fs *helloFS) ReadDir(
 
 	// Grab the range of interest.
 	if op.Offset > fuseops.DirOffset(len(entries)) {
-		return fuse.EIO
+		return nil
 	}
 
 	entries = entries[op.Offset:]

samples/memfs/callback_test.go

@@ -0,0 +1,81 @@
+package memfs_test
+
+import (
+	"os"
+	"path"
+	"testing"
+
+	"github.com/jacobsa/fuse/samples"
+	"github.com/jacobsa/fuse/samples/memfs"
+	. "github.com/jacobsa/ogletest"
+)
+
+func TestFuseServerCallbacks(t *testing.T) { RunTests(t) }
+
+type CallbackTest struct {
+	samples.SampleTest
+	readFileCallbackInvoked  bool
+	writeFileCallbackInvoked bool
+}
+
+func init() { RegisterTestSuite(&CallbackTest{}) }
+
+func (t *CallbackTest) SetUp(ti *TestInfo) {
+	t.MountConfig.DisableWritebackCaching = true
+	t.readFileCallbackInvoked = false
+	t.writeFileCallbackInvoked = false
+
+	t.Server = memfs.NewMemFSWithCallbacks(
+		currentUid(),
+		currentGid(),
+		func() { t.readFileCallbackInvoked = true },
+		func() { t.writeFileCallbackInvoked = true },
+	)
+	t.SampleTest.SetUp(ti)
+}
+
+// The test suite is torn down during the test to ensure
+// that all FUSE operations are complete before checking
+// the invocations on the callbacks.
+func (t *CallbackTest) TearDown() {}
+
+func (t *CallbackTest) TestCallbacksInvokedForWriteFile() {
+	AssertEq(t.writeFileCallbackInvoked, false)
+	AssertEq(t.readFileCallbackInvoked, false)
+
+	// Write a file.
+	fileName := path.Join(t.Dir, memfs.CheckFileOpenFlagsFileName)
+	const contents = "Hello world"
+	err := os.WriteFile(fileName, []byte(contents), 0400)
+	AssertEq(nil, err)
+
+	// Tear down the FUSE mount. This ensures that all FUSE operations are complete.
+	t.SampleTest.TearDown()
+
+	// Check that our callback was invoked as expected.
+	AssertEq(t.writeFileCallbackInvoked, true)
+	AssertEq(t.readFileCallbackInvoked, false)
+}
+
+func (t *CallbackTest) TestCallbacksInvokedForWriteFileAndReadFile() {
+	AssertEq(t.writeFileCallbackInvoked, false)
+	AssertEq(t.readFileCallbackInvoked, false)
+
+	// Write a file.
+	fileName := path.Join(t.Dir, memfs.CheckFileOpenFlagsFileName)
+	const contents = "Hello world"
+	err := os.WriteFile(fileName, []byte(contents), 0400)
+	AssertEq(nil, err)
+
+	// Read it back.
+	slice, err := os.ReadFile(fileName)
+	AssertEq(nil, err)
+	ExpectEq(contents, string(slice))
+
+	// Tear down the FUSE mount. This ensures that all FUSE operations are complete.
+	t.SampleTest.TearDown()
+
+	// Check that our callback was invoked as expected.
+	AssertEq(t.writeFileCallbackInvoked, true)
+	AssertEq(t.readFileCallbackInvoked, true)
+}

samples/memfs/inode.go

@@ -17,6 +17,7 @@ package memfs
 import (
 	"fmt"
 	"io"
+	"io/fs"
 	"os"
 	"time"
 
@@ -377,7 +378,8 @@ func (in *inode) SetAttributes(
 
 	// Change mode?
 	if mode != nil {
-		in.attrs.Mode = *mode
+		in.attrs.Mode &= ^fs.ModePerm
+		in.attrs.Mode |= *mode & fs.ModePerm
 	}
 
 	// Change mtime?

samples/memfs/memfs.go

@@ -67,6 +67,9 @@ type memFS struct {
 	// INVARIANT: This is all and only indices i of 'inodes' such that i >
 	// fuseops.RootInodeID and inodes[i] == nil
 	freeInodes []fuseops.InodeID // GUARDED_BY(mu)
+
+	readFileCallback  func()
+	writeFileCallback func()
 }
 
 // Create a file system that stores data and metadata in memory.
@@ -77,11 +80,21 @@ type memFS struct {
 func NewMemFS(
 	uid uint32,
 	gid uint32) fuse.Server {
+	return NewMemFSWithCallbacks(uid, gid, nil, nil)
+}
+
+func NewMemFSWithCallbacks(
+	uid uint32,
+	gid uint32,
+	readFileCallback func(),
+	writeFileCallback func()) fuse.Server {
 	// Set up the basic struct.
 	fs := &memFS{
-		inodes: make([]*inode, fuseops.RootInodeID+1),
-		uid:    uid,
-		gid:    gid,
+		inodes:            make([]*inode, fuseops.RootInodeID+1),
+		uid:               uid,
+		gid:               gid,
+		readFileCallback:  readFileCallback,
+		writeFileCallback: writeFileCallback,
 	}
 
 	// Set up the root inode.
@@ -627,7 +640,7 @@ func (fs *memFS) OpenFile(
 		// Set attribute (name=fileOpenFlagsXattr, value=OpenFlags) to test whether
 		// we set OpenFlags correctly. The value is checked in test with getXattr.
 		value := make([]byte, 4)
-		binary.LittleEndian.PutUint32(value, uint32(op.OpenFlags))
+		binary.LittleEndian.PutUint32(value, uint32(op.OpenFlags)&syscall.O_ACCMODE)
 		err := fs.setXattrHelper(inode, &fuseops.SetXattrOp{
 			Name:  FileOpenFlagsXattrName,
 			Value: value,
@@ -653,6 +666,8 @@ func (fs *memFS) ReadFile(
 	var err error
 	op.BytesRead, err = inode.ReadAt(op.Dst, op.Offset)
 
+	op.Callback = fs.readFileCallback
+
 	// Don't return EOF errors; we just indicate EOF to fuse using a short read.
 	if err == io.EOF {
 		return nil
@@ -673,6 +688,8 @@ func (fs *memFS) WriteFile(
 	// Serve the request.
 	_, err := inode.WriteAt(op.Data, op.Offset)
 
+	op.Callback = fs.writeFileCallback
+
 	return err
 }
 

samples/mount_hello/mount.go

@@ -50,6 +50,7 @@ func main() {
 
 	cfg := &fuse.MountConfig{
 		ReadOnly: *fReadOnly,
+		FuseImpl: fuse.FUSEImplMacFUSE,
 	}
 
 	if *fDebug {

samples/roloopbackfs/inode.go

@@ -21,7 +21,7 @@ import (
 	"path/filepath"
 	"sync"
 	"sync/atomic"
-	"time"
+	"syscall"
 
 	"github.com/jacobsa/fuse/fuseops"
 	"github.com/jacobsa/fuse/fuseutil"
@@ -53,21 +53,20 @@ func getOrCreateInode(inodes *sync.Map, parentId fuseops.InodeID, name string) (
 		return nil, nil
 	}
 	parentPath := parent.(Inode).Path()
-	entries, err := ioutil.ReadDir(parentPath)
+
+	path := filepath.Join(parentPath, name)
+	fileInfo, err := os.Stat(path)
 	if err != nil {
-		return nil, err
+		return nil, nil
 	}
-	for _, entry := range entries {
-		if entry.Name() == name {
-			inodeEntry := &inodeEntry{
-				id:   nextInodeID(),
-				path: filepath.Join(parentPath, name),
-			}
-			storedEntry, _ := inodes.LoadOrStore(inodeEntry.id, inodeEntry)
-			return storedEntry.(Inode), nil
-		}
+	stat, _ := fileInfo.Sys().(*syscall.Stat_t)
+
+	inodeEntry := &inodeEntry{
+		id:   fuseops.InodeID(stat.Ino),
+		path: path,
 	}
-	return nil, nil
+	storedEntry, _ := inodes.LoadOrStore(inodeEntry.id, inodeEntry)
+	return storedEntry.(Inode), nil
 }
 
 type inodeEntry struct {
@@ -101,16 +100,14 @@ func (in *inodeEntry) Attributes() (*fuseops.InodeAttributes, error) {
 	if err != nil {
 		return &fuseops.InodeAttributes{}, err
 	}
+
 	return &fuseops.InodeAttributes{
-		Size:   uint64(fileInfo.Size()),
-		Nlink:  1,
-		Mode:   fileInfo.Mode(),
-		Atime:  fileInfo.ModTime(),
-		Mtime:  fileInfo.ModTime(),
-		Ctime:  time.Now(),
-		Crtime: time.Now(),
-		Uid:    uid,
-		Gid:    gid,
+		Size:  uint64(fileInfo.Size()),
+		Nlink: 1,
+		Mode:  fileInfo.Mode(),
+		Mtime: fileInfo.ModTime(),
+		Uid:   uid,
+		Gid:   gid,
 	}, nil
 }
 
@@ -119,12 +116,12 @@ func (in *inodeEntry) ListChildren(inodes *sync.Map) ([]*fuseutil.Dirent, error)
 	if err != nil {
 		return nil, err
 	}
-	dirents := make([]*fuseutil.Dirent, len(children))
+	dirents := []*fuseutil.Dirent{}
 	for i, child := range children {
 
 		childInode, err := getOrCreateInode(inodes, in.id, child.Name())
 		if err != nil || childInode == nil {
-			return nil, nil
+			continue
 		}
 
 		var childType fuseutil.DirentType
@@ -136,12 +133,12 @@ func (in *inodeEntry) ListChildren(inodes *sync.Map) ([]*fuseutil.Dirent, error)
 			childType = fuseutil.DT_File
 		}
 
-		dirents[i] = &fuseutil.Dirent{
+		dirents = append(dirents, &fuseutil.Dirent{
 			Offset: fuseops.DirOffset(i + 1),
 			Inode:  childInode.Id(),
 			Name:   child.Name(),
 			Type:   childType,
-		}
+		})
 	}
 	return dirents, nil
 }

samples/roloopbackfs/roloopbackfs.go

@@ -121,7 +121,7 @@ func (fs *readonlyLoopbackFs) ReadDir(
 	}
 
 	if op.Offset > fuseops.DirOffset(len(children)) {
-		return fuse.EIO
+		return nil
 	}
 
 	children = children[op.Offset:]
@@ -139,8 +139,13 @@ func (fs *readonlyLoopbackFs) ReadDir(
 func (fs *readonlyLoopbackFs) OpenFile(
 	ctx context.Context,
 	op *fuseops.OpenFileOp) error {
-	// Allow opening any file.
+
+	var _, found = fs.inodes.Load(op.Inode)
+	if !found {
+		return fuse.ENOENT
+	}
 	return nil
+
 }
 
 func (fs *readonlyLoopbackFs) ReadFile(

samples/statfs/statfs_darwin_test.go

@@ -16,6 +16,7 @@ package statfs_test
 
 import (
 	"fmt"
+	"github.com/jacobsa/oglematchers"
 	"math"
 	"regexp"
 	"syscall"
@@ -68,7 +69,7 @@ func (t *StatFSTest) Syscall_ZeroValues() {
 	ExpectEq(0, stat.Bavail)
 	ExpectEq(0, stat.Files)
 	ExpectEq(0, stat.Ffree)
-	ExpectEq("osxfuse", convertName(stat.Fstypename[:]))
+	ExpectThat(convertName(stat.Fstypename[:]), oglematchers.AnyOf(oglematchers.Equals("osxfuse"), oglematchers.Equals("macfuse")))
 	ExpectEq(t.canonicalDir, convertName(stat.Mntonname[:]))
 	ExpectEq(fsName, convertName(stat.Mntfromname[:]))
 }
@@ -103,7 +104,7 @@ func (t *StatFSTest) Syscall_NonZeroValues() {
 	ExpectEq(canned.BlocksAvailable, stat.Bavail)
 	ExpectEq(canned.Inodes, stat.Files)
 	ExpectEq(canned.InodesFree, stat.Ffree)
-	ExpectEq("osxfuse", convertName(stat.Fstypename[:]))
+	ExpectThat(convertName(stat.Fstypename[:]), oglematchers.AnyOf(oglematchers.Equals("osxfuse"), oglematchers.Equals("macfuse")))
 	ExpectEq(t.canonicalDir, convertName(stat.Mntonname[:]))
 	ExpectEq(fsName, convertName(stat.Mntfromname[:]))
 }