|
First, I think you should upgrade to the lastest
CVS head (joke, of course).
Some informations you have to know about
lwIP:
- lwIP 1.2.0 - and even lastest cvs head - are not
safe-thread about using two application threads on a same socket
- lwIP 1.2.0 tcp close was bugged, and is just
fixed on CVS head
But, even if you don't upgrade to CVS head,
adding recv timeouts on your code is not very difficult, but in this case, this
is not a solution. A first patch should be to add a "sys_mbox_post(telnetNetCon->recvmbox, NULL);" before the netconn_close
(perhaps after that with a little vTaskDelay). Like this, the thread blocked on
netconn_recv will be directly wake-up.
----- Original Message -----
Sent: Tuesday, October 30, 2007 6:37
PM
Subject: [lwip-users] Telnet server/LWIP
crash
This is a lengthy explanation, but after 2-3 weeks of banging my head
into the wall I think it is necessary.
I am working on a Telnet server using 1.2.0 release (as I am unable to
get latest CVS head to compile and run on the AT91SAM7X demo board under
FreeRTOS). Because 1.2.0 does not support receive timeouts, the server
is implemented as two tasks. One handles the "receive from network" and
the other "send to network" functions, using FreeRTOS queues to send character
data to/from the application task.
Before you say "upgrade to the latest CVS head" - I have already tried
unsuccessfully for much of this time to do just that, however the update
simply doesn't work, even when I roll back to just porting the new LWIP to the
original FreeRTOS demo that I got with the original.
A little more background...
It is possible for the application task to request a
session end (primarily to log out idle users). This is accomplished with
the following code:
static void
close_telnet (void)
{
if (net_connection_open)
{
// Network connection is closed again
net_connection_open = FALSE;
// Don't do this again until
new connection starts
idletimer = 0;
netconn_close (telnetNetCon);
tn_isonline[0] =
FALSE;
while (netconn_delete (telnetNetCon) !=
ERR_OK)
{
vTaskDelay (TELNET_SHORT_DELAY);
}
vTaskDelay
(TELNET_LONG_DELAY);
}
}
(Note: The variable "net_connection_open" is a flag which is set TRUE
when the receiver task gets a connection from a client and starts
communicating with them.)
While
this is going on, the "read from network" task is waiting in a netconn_recv()
call. It does not return immediately when the other task calls
close_telnet() above (which I would expect due to the netconn_close() and
netconn_delete() calls).
The
client tells me immediately that it has lost its connection when
close_telnet() is executed.
If I
then attempt another connection from a client, the "read from network" task
returns from the netconn_recv() it has been in, but the return value is
pointing to the same location as when it was receiving valid data. At
this point I just netbuf_delete() it, then go back to a netconn_accept()
waiting for the next telnet client to attach.
It accepts the connection, returning same pointer
address as previous connection. Starts the connection process (which
includes sending a DO and DONT string to the client), but when the program
starts attempting to wait for more data to come from the new client, the micro
resets with a data abort (on the ARM7, this generally means that data was
either mis-aligned or not present).
Another scenario - the client disconnects from the server. When
this happens, I am able to start a new client and log in once successfully,
but logging out and attempting a third time results in a data abort
also.
If you need the code, I
would be happy to provide it. If you have ANY advice, I'd love to hear
it.
Thanks.
Andy
_______________________________________________
lwip-users mailing
list
address@hidden
http://lists.nongnu.org/mailman/listinfo/lwip-users
|