Re: [lwip-users] portscan

[Bernhard 'Gustl' Bauer]

address@hidden schrieb:
> Bernhard 'Gustl' Bauer wrote:
> > Simon Goldschmidt schrieb:
> >  
> > > In the past, stalls like this have often been driver bugs: a portscan
> > > is like a stress situation to your driver, maybe it leaves interrupts
> > > disabled or something like that, so the device cannot receive any
> > > packets?
> > >     
> >
> > I enabled the stats of my driver (CS8900). What I get during a 
> > portscan is a lot of misses and collisions. Could this crash TCP or ping?
> >
> >   
>
> I shouldn't crash TCP: if a packet is not received correctly by a remote 
> host, TCP retransmits it eventualy, so your connections should go back 
> to normal after the portscan has finished. During the portscan, however, 
> it could well be that your device is not fast enough to process all the 
> packets sent by the PC, so TCP connections might stall and ping might 
> fail while the portscan is active. That's perfectly normal for a device 
> that cannot handle received packets as fast as they can be sent on the 
> wire (which most embedded devices can't).

I use DMA to transfer packets from CS8900 to a RAM buffer. There can be
a problem if the buffer is filled faster than processed by LWIP. I tried 
to avoid this by dropping packets that have been already received from 
the CS8900. This was disastrous. Nothing worked anymore. I drew two 
conclusions out of this:
- a ethernet tx throttles back if the connected rx has misses
- a packet received by a rx has to be processed under all circumstances

Please correct me when I'm wrong.

My next step would be to delay the DMAs if there is no space in the buffer.

Gustl