mount: address review on passthrough

- Move teardownPassthrough() ahead of the fhLockTable lock in ReleaseHandle (it is self-synchronized by passthroughMu and needs no other lock), removing any lock-order coupling between passthroughMu and fhLockTable.

- Reuse wfs.copyBufferPool in materializeFile instead of allocating a fresh buffer per materialization.

weed/mount/filehandle.go

@@ -153,6 +153,12 @@ func (fh *FileHandle) AddChunks(chunks []*filer_pb.FileChunk) {
 }
 
 func (fh *FileHandle) ReleaseHandle() {
+	// Tear down FUSE passthrough first, outside the fhLockTable lock.
+	// teardownPassthrough is self-synchronized by passthroughMu and needs no
+	// other lock, so keeping it out of the fhLockTable critical section avoids
+	// any lock-order coupling between passthroughMu and fhLockTable.
+	fh.teardownPassthrough()
+
 	// Release distributed lock before cleaning up, so other mounts can
 	// proceed as soon as this handle is done flushing.
 	if fh.dlmLock != nil {
@@ -165,7 +171,6 @@ func (fh *FileHandle) ReleaseHandle() {
 	defer fh.wfs.fhLockTable.ReleaseLock(fh.fh, fhActiveLock)
 
 	fh.dirtyPages.Destroy()
-	fh.teardownPassthrough()
 	if IsDebugFileReadWrite {
 		fh.mirrorFile.Close()
 	}

weed/mount/weedfs_passthrough_linux.go

@@ -111,7 +111,10 @@ func (wfs *WFS) setupPassthroughBacking(fh *FileHandle, fileSize int64) (int32,
 // path the kernel Read handler uses, so the backing bytes are identical to what
 // a normal read would return.
 func materializeFile(fh *FileHandle, dst *os.File, fileSize int64) error {
-	buf := make([]byte, 1024*1024)
+	// Reuse the shared copy buffer pool (ChunkSizeLimit-sized) rather than
+	// allocating a fresh buffer per materialization.
+	buf := fh.wfs.copyBufferPool.Get().([]byte)
+	defer fh.wfs.copyBufferPool.Put(buf)
 	for offset := int64(0); offset < fileSize; {
 		// readDataByFileHandle reads chunks + dirty pages and maps EOF to nil.
 		n, err := readDataByFileHandle(buf, fh, offset)