RE: [lwip-users] Tcp freeze

[Dave Perry]

Hi all,

I found the problem and a solution, but it raises a question.

The problem was in my implementation of sys_mbox_post. lwIP is supposed to
work with queue sizes as small as one, and this is what I was using. My
implementation would drop a message instead of blocking if the queue was
full. 

The problem occurred when two SYN packets were received one after another.
The application layer would never hear about the second connection because
of the lost message. Blocking in sys_mbox_post was not enough to solve the
problem, I also had to increase the queue size from 1 to 10.

Is this a case where lwIP will not work properly with a queue size of 1?

Dave


-----Original Message-----
From: address@hidden
[mailto:address@hidden On Behalf Of Dave
Perry
Sent: Tuesday, July 19, 2005 9:53 AM
To: 'Mailing list for lwIP users'
Subject: [lwip-users] Tcp freeze

I've been reluctant to bring this up because it could easily be a bug in my
new port of lwIP, but I think it could be in lwIP.

I have a test app that serves a web page, and the page reloads every second.
I open several windows and allow them to keep requesting pages. After a few
seconds, the system freezes and no more packets are sent from lwIP.
Sometimes it recovers and starts sending again, but in this case it will
have leaked a few pbufs.

If I define MEMP_NUM_TCP_PCB to 1, the problem goes away completely. So
maybe it's something to do with allocating pcbs? I have an ethereal dump if
anybody is interested.


Dave Perry



_______________________________________________
lwip-users mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/lwip-users