Re: [lwip-users] Problem with ftp session

[Jonathan Larmour]

Thomas Taranowski wrote:
> 95% of the time my ftp/telnet apps work fine.  However, once in awhile I
> get into a cycle where the tcp connections totally hose up.  The cause
> appears to be related to the fact that in sys_mbox_fetch(), the
> again:/goto again loop is spinning with wild abandon.  I put a counter
> inside sys_mbox_fetch, which easily upticks into the several hundreds
> within seconds of stack startup.  I've verified that the
> sys_arch_mbox_fetch is returning sane values.  The odd thing is that
> this only crops up once in awhile, and will stay around.  Last time I
> thought I had a bad ethernet card, and switched test targets, and
> everything started working.  However, now that I'm seeing it again on
> different hardware/ethernet card combo, I'm concerned that I've got a
> critical bug floating around somewhere.
> 
> I've attached a wireshark capture of the link output.  It's a portion of
> the ftp session.  You'll see that it basically works, but will spew out
> loads of TCP Dup ACK, TCP Retransmits, and finally drops the connection
> at the end with a series of 7 RSTs in a row.  Presumably I'm seeing
> these cause the timers are firing way to fast.
> 
> Note: This is a custom port of lwIP to an OS I'm working on.
> 
> Any ideas on how to debug this, or what the issue is?  When this occurs,
> I do have an inetd server start listening on several tcp ports at system
> startup.

Given sys_mbox_fetch is spinning, I would initially suspect your port's
sys_arch_timeouts() call. It should return a list of timeouts only for the
current thread.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["Si fractum non sit, noli id reficere"]------       Opinions==mine