FTPSRV_release

[yb]

CW 10.6 / MQX 4.2 / K60

I'm using FTPSRV with no problems, everything is OK... except if I have an active connection and then, call FTPSRV_release().

Since this call, EVERY tasks on my projects are blocked for ever in ftpsrc.c line 96 :

    while(server->server_tid)

    {

        _sched_yield() ;

    }

I try to add a _time_delay() in this loop and I have other tasks OK. But the FTPSRV NEVER shutdown before I cut the FTP client.

Nobody shutdown the FTPSRV() ?

Yvan

[stefanhaase]

The call to FTPSRV_release()  may get stuck also because of a wrong task priority schema. The RTCS user manual gives a hint about it.

I got the case with following task priorities:

 TCP_TASK  =  10

 FTP_SERVER_TASK = 7

 TASK that calls FTPSRV_release() = 8

In the manual it's stated that FTP_SERVER_TASK must be lower than TCP_TASK. I would like to add, also the call task of FTPSRV_release() must be lower. So after adjusting the task priorities to:

 TCP_TASK  =  10

 FTP_SERVER_TASK = 11

 TASK that calls FTPSRV_release() = 12

FTPSRV exits gracefully.

Best Regards,

Stefan Haase

[Fabi]

This thread is more than one year old. But I see same effect in MQX 4.2.0 and no fix could be found.

A work around which works for me is to use following code:

        FTPSRV_STRUCT*  server = (FTPSRV_STRUCT*)FTPs_hd;
        result = shutdown ( server->sock_v4, FLAG_ABORT_CONNECTION );

instead of calling FTPSRV_release().

It looks like that closesocket() does not release the select() in ftpsrv_wait_for_conn() but shutdown() does.

[Martin_]

The release mechanism depends on the ftpsrv_server_task() setting the server->server_tid member to zero. As it is a different task than the one that calls FTPSRV_release(), I think it must have "volatile" keyword to force the compiler to always read from memory when checking the value.

Maybe your compiler optimized this out - you can check by looking into assembly when inside the while loop. Try to declare "server_tid" as volatile.

[yb]

Martin,

Thanks for your answer, but my problem is not with a particular use : server_tid is the one which is include in the Freescale FTPSRV_STRUCT struct.

I've just tried your idea, but there is no difference.

I think that everybody who try to use the FTPSRV will have this symptom...

Yvan