[FIXED] MQTT: Possible panic when clients misbehave

If a client with a given client ID is connected and while connected
another client tries to reuse the same client ID, the spec says that
the old client be closed and the new one accepted.
However, the server protects from this flapping happening all the time
by rejecting new clients that try to connect at a very fast pace.

However, the server was closing a misbehaving client after a second
delay (to prevent immediate reconnect if the client library does that)
but was not blocking the read loop and the compounding issue was that
if that misbehaving client is REALLY misbehaving and not waiting for
the CONNACK to send more protocols (for instance SUB) the server would
panic because the client was not fully configured.

To prevent that, the server will now "block" this misbehaving client
in its readLoop before closing the connection, preventing processing
of possible protocols that follow the CONNECT.

Resolves #3313

Signed-off-by: Ivan Kozlovic <ivan@synadia.com>

server/mqtt_test.go

@@ -4361,8 +4361,22 @@ func TestMQTTFlappingSession(t *testing.T) {
 	defer c.Close()
 	proto := mqttCreateConnectProto(ci)
 	if _, err := testMQTTWrite(c, proto); err != nil {
-		t.Fatalf("Error writing connect: %v", err)
+		t.Fatalf("Error writing protocols: %v", err)
 	}
+	// Misbehave and send a SUB protocol without waiting for the CONNACK
+	w := &mqttWriter{}
+	pkLen := 2 // for pi
+	// Topic "foo"
+	pkLen += 2 + 3 + 1
+	w.WriteByte(mqttPacketSub | mqttSubscribeFlags)
+	w.WriteVarInt(pkLen)
+	w.WriteUint16(1)
+	w.WriteBytes([]byte("foo"))
+	w.WriteByte(1)
+	if _, err := testMQTTWrite(c, w.Bytes()); err != nil {
+		t.Fatalf("Error writing protocols: %v", err)
+	}
+	// Now read the CONNACK and we should have been disconnected.
 	if _, err := testMQTTRead(c); err == nil {
 		t.Fatal("Expected connection to fail")
 	}