packet unmarshalling code into main server loop

This means we are now passing full packets to the workers. This makes
the request id's available which is required for fixing packet ordering.
Benchmark tests weren't affected at all.

server.go

@@ -29,7 +29,7 @@ type Server struct {
 	serverConn
 	debugStream   io.Writer
 	readOnly      bool
-	pktChan       chan rxPacket
+	pktChan       chan packet
 	openFiles     map[string]*os.File
 	openFilesLock sync.RWMutex
 	handleCount   int
@@ -83,7 +83,7 @@ func NewServer(rwc io.ReadWriteCloser, options ...ServerOption) (*Server, error)
 			},
 		},
 		debugStream: ioutil.Discard,
-		pktChan:     make(chan rxPacket, sftpServerWorkerCount),
+		pktChan:     make(chan packet, sftpServerWorkerCount),
 		openFiles:   make(map[string]*os.File),
 		maxTxPacket: 1 << 15,
 	}
@@ -123,12 +123,7 @@ type rxPacket struct {
 
 // Up to N parallel servers
 func (svr *Server) sftpServerWorker() error {
-	for p := range svr.pktChan {
-
-		pkt, err := makePacket(p)
-		if err != nil {
-			return err
-		}
+	for pkt := range svr.pktChan {
 
 		// readonly checks
 		readonly := true
@@ -297,14 +292,22 @@ func (svr *Server) Serve() error {
 	}
 
 	var err error
+	var pkt packet
 	var pktType uint8
 	var pktBytes []byte
 	for {
 		pktType, pktBytes, err = svr.recvPacket()
 		if err != nil {
 			break
 		}
-		svr.pktChan <- rxPacket{fxp(pktType), pktBytes}
+
+		pkt, err = makePacket(rxPacket{fxp(pktType), pktBytes})
+		if err != nil {
+			svr.conn.Close() // shuts down recvPacket
+			break
+		}
+
+		svr.pktChan <- pkt
 	}
 
 	close(svr.pktChan) // shuts down sftpServerWorkers