Re: RE : [lwip-users] netconn_connct returning always ERR_OK

[Marko Panger]

Hi Jonathan,

Good point !! Now I remember that something was very strange to me 
regarding this semaphore. Well, to be sure I have to check with the 
debugger but for now I can only comment what could be happening inside. 
I'll try to describe it:

1) netconn_new_with_proto_and_callback() creates the semaphore and the 
semaphore is initialized to 0. After that tcpip_apimsg() is invoked 
which sends the do_newconn() function to the main tcp thread. The 
tcpip_apimsg() function pends on the op_completed semaphore with 0 time.

2) Depending on the task priorities the tcpip_apimsg() function might 
return immediately with ERR_OK. When the main tcp thread completes the 
do_newconn() function the op_semaphore is posted and after that it has 
the value of 1.

3) When then I call netconn_connect() the semaphore is already posted, 
thus returning immediately with ERR_OK.

This is what I can reduce from observing the code. It seems like a race 
condition which is depends on task priorities. Please don't get me wrong 
- I have to check this with the debugger.

marko


Jonathan Larmour wrote:
> Marko Panger wrote:
> > Hi again,
> >
> > I'm still unable to reproduce the proper behavior or the behavior as 
> > Frédéric described below. The point is still that netconn_connect() 
> > returns ERR_OK even if there is nobody listening on the other side.
>
> What happens behind the scenes is that a message will get posted to 
> the tcpip thread's mailbox, and the calling thread should wait on a 
> per-connection semaphore (called op_completed) for the tcpip thread to 
> indicate that the operation has completed.
>
> Perhaps somehow that semaphore did not have the correct state before 
> the call - perhaps it had already been signaled? Or had not been 
> created in the port with a value of 0 - in netconn_alloc() in 
> api_msg.c, lwIP requests it to be created with a value of 0, but 
> perhaps something doesn't do it right in the port?
>
> Jifl