[lwip-devel] communication hangs after single ACK

[Albert Bartel]

Hello lwip-developers,

 

after having some problems with an Atmel UC3 port of lwip-stack, I fixed them with some help of some lwip developers. There was a big bug in the port so that the ethernetif_input task called ethernet_input() directly instead of calling  netif->input(). So it was not thread safe.

After fixing this bug, I also updated the stack to the lates version (V1.3.2 RC1) and hoped that the strange behaviour would be fixed, since there have been some changes in the stack.

But the problem still occurs.

 

I will try to describe the problem:

A colleague has got a PC that runs Windows Vista and is connected over a WLAN-Router (signal strength may be not good).

I got a PC that runs Windows XP and is connected directly to a Router (not WLAN).

The Server is a small Atmel UC3A Webserver that uses lwip-stack and supports only HTTP/1.0 (no keep alive), so for each request a new connection has to be established.

With my PC everything works fine, I got no problems. But with the PC of my colleague it often occurs, that the PC receives no data from the client while the connection will stay opened (ESTABLISHED).

I attached a small Ethereal trace with a request when this situation happened.

 

It is like that:

PC                   | Server

--------------------------

SYN                 |

                        | SYN,ACK

ACK                 |

HTTP GET        |

                        | ACK

Connection hangs, no response data for this request is sent.

FIN, ACK          |

                        | ACK

After manually closing the connection (after more that 60s) the data that should have been sent as response is sent - but after the connection has been closed.

                        | HTTP OK with data

RST, ACK         |

                        | FIN, ACK

 

Other requests from the same PC to the webserver, while this connection hangs, are handled and answered correctly.

 

If everything works fine, directly after sending the GET message I get a HTTP/1.0 200 OK message (with ACK for the GET), but in the cases, when I get no data and the connection stays open, I only get an ACK for the GET, but no data follows. The data comes in a separate packet after closing the connection manually.

 

The GET must be seen by the stack because of the correct ACK, but somehow the request is not processed by the webserver. After the FIN, ACK from the client recv_tcp() sets the api_event NETCONN_EVT_RCVPLUS and the webserver processes the GET after lwip_selscan returns >0.

 

If I turn on TCP_INPUT_DEBUG  and TCP_OUTPUT_DEBUG, the debug output slows down the whole system and this problem does not occur anymore.

 

Is it possible, that an api_event (NETCONN_EVT_RCVPLUS/ NETCONN_EVT_SENDPLUS) somehow gets lost?

Have you got an idea or a hint, why this happens?

Many thanks in advance.

 

Greetings

Albert

 

communication_hangs.pcap
Description: Binary data