[lwip-users] strange tcp connection issue

[Thomas Taranowski]

On my platform I have an ftp server listening on port 21.  During
automated testing, where the ftp is accepting and closing connections
frequently, the target will sometimes stop listening on port 21 for no
apparent reason.  All other apps listening on other ports still
function correctly.  I've debugged it down to where I find a
discrepancy between the socket layer and the stack:

I have a real time display which shows the state of all sockets
current active in the system.  At the time of the problem, this
correctly shows a TCP socket listening on port 21.  To get this data,
I dereference the socket->conn->pcb.tcp->state variable in the user
application context.  I also added some code to tcp_input to print out
the tcp_listen_pcb list if no matching PCB was found for the incoming
segment (down by where it calls tcp_rst).  What I found was
interesting.  I see something like the following:
  listening on port: 1030
  listening on port: 1027
    ...
  listening on port: 0   <-- this one WAS 21, but now isn't.

Any ideas on what might cause such a discrepancy?  This is a late 1.2 cvs head.



-- 
Thomas Taranowski
Certified netburner consultant
baringforge.com