Re: [ECOS] Problems when aborting tftp transfers

["Øyvind Harboe" <oyvind.harboe@zylin.com>]

On Tue, 2004-08-03 at 11:49, Andrew Lunn wrote:
> On Tue, Aug 03, 2004 at 08:40:46AM +0200, ?yvind Harboe wrote:
> > I haven't got the setup available right now, but I think I remember the steps 
> > to reproduce a bit better:
> > 
> > 1. start tftp transfer
> > 2. abort tftp transfer
> > 3. at this point the new tftp server thread is waiting for timeout on
> > #1
> > 4. start new tftp transfer
> > 5. wait a bit(but not long enough for tftp transfer to start).
> > 6. abort tftp transfer
> > 7. at this point the tftp server is stuck in the main loop timing out on
> > WRQ requests. Each WRQ request requires a tftpd_write_file() to time out, so
> > this takes a loooong time.
> 
> This now make sense. It is the starting of the new connection before
> the old one times out that causes the delay. Since the system is
> single threaded it has two choices when the new connection request
> arrives. The current implementation simply buffers the request until
> later. The alternative implementation would be to discard the request
> and let the client retry. Both implementations are probably correct
> but have pros and cons. All these aborted connections is a corner case
> that should not happen very often so i would not worry about it.
> 
> Having said that, i think the real solution is to teach your users to
> have some patients and not abort the connections....

:-)

The problem is that the timeout is *very* long because it is number of 
queued WRQ (ca 5) * timeout for tftpd_write_file().

A single timeout is acceptable, but minutes is very confusing for the
user.

>         Andrew

-- 
Øyvind Harboe
http://www.zylin.com



--
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss