Re: [MiNT] tcp_select ( was: Re: inet4: errno is not set correct )

[Alan Hourihane <alanh@fairlite.co.uk>]

On Mon, 2010-10-04 at 21:32 +0200, m0n0 wrote:
> Am Sonntag, den 03.10.2010, 14:42 +0200 schrieb m0n0
> <ole@monochrom.net>:
> 
> > Anyone is able to fix this, or maybe someone knows 
> > more about the network code and can acknowledge this 
> > bug at source level? 
> 
> I think this could be fixed in tcp_select... 
> 
> here is part of tcp_select function:
> 			switch (tcb->state)
> 			{
> 				case TCBS_SYNSENT:
> 				case TCBS_ESTABLISHED:
> 					break;
> 				default:
> 					return 1;
> 			}
> 			
> 			return so_rselect (so, proc);
> 
> Intentionally I would try to return 0 when tcb->state == TCBS_SYNSENT, 
> at this time I just don't have the knowledge to build my own kernel ;) (
> But I think I will dig into it because of this one..) 
> 
> Another problem I would have with that change: Is tcp_select only used
> by user space applications? If it is also used by the kernel during TCP
> handshake, then the TCBS_SYNSENT state of course must return 1... but I
> guess it's not meant to be used within the kernel... 
> 
> On the other hand, I believe return 0 when tcb->state == TCBS_SYNSENT
> won't be enough... TCBS_SYNRCVD is still handled by default and will
> return 1, which seems to be wrong from my point of view. 
> 
> Anybody wants to talk about this =) ?

Sorry Ole, been busy.

It's always useful to come up with a small test case that demonstrates
the problem.

Can you do that and I'll be sure to look at it.

Alan.