Re: [lwip-users] tcpip thread gets stuck when socket is closed

[Kieran Mansley]

On Sat, 2011-09-17 at 15:51 +0200, Stefan Lankes wrote:
> 
> > 
> > On 15 Sep 2011, at 15:43, Stefan Lankes wrote:
> > 
> >> we use LwIP 1.4.0 as TCP/IP stack for an internal OS project. We have 
> >> already tested some applications, which used the LwIP's Socket API - they 
> >> work fine. However, if the applications close the socket, the tcpip thread 
> >> gets stuck and isn't able to handle incoming messages. To analyze this 
> >> behavior, we defined the macro LWIP_TCPIP_THREAD_ALIVE to print some 
> >> messages periodically on the screen. This stops by closing the socket. 
> >> 
> >> Do you have any idea, why the Socket API doesn't work correctly?
> > 
> > You've probably violated lwIP's threading constraints and corrupted some 
> > internal state such that the tcpip thread gets stuck in a loop.  If you can 
> > describe all the different threads you have and how they interface to lwIP 
> > that would probably reveal the problem.
> > 
> > Kieran
> 
> I read that the some users has similar problems. Most of these users
> violated the LwIP's threading constraints. Therefore, I created only
> one thread which realized a simple socket-based server. Like the Unix
> port, the tcpip thread initialized in its callback the NICs. Beside
> these threads, my system has also an idle thread and an initd thread,
> which creates my server example and terminates.

How does your device driver pass packets to lwIP?  There have been
problems in the past where people have not used that interface
correctly.

> On a real machine, the tcpip thread stucks when the socket is closed.
> I tested my code also on qemu. By closing a socket, I got the
> assertion "men_free: legal memory failed at line 320 in
> lwip/src/core/mem.c".

That suggests that something is trying to free an invalid pointer.
Again this points at state corruption.  I wonder if you have problems
with stack overflow on your system or something like that?

Kieran