Re: [PATCH] ehea: Fix a checksum issue on the receive path

[Eric Dumazet <eric.dumazet@gmail.com>]

Le vendredi 08 octobre 2010 à 11:14 -0300, Breno Leitao a écrit :
> Hi Eric
> 
> On 10/08/2010 01:45 AM, Eric Dumazet wrote:
> > Just to be clear : packets with wrong checksums are not given to upper
> > stack, so a tcpdump can not display them ? I am not sure many drivers do
> > that.
> Well, what my code does is: 1) if the current packet is a UDP/TCP, then 
> the checksum is not necessary, since we would check the checksum on 
> ehea_proc_rwqes(), specific at this part of the code:
> 
>                 if (!ehea_check_cqe(cqe, &rq)) {
> 			// Send the packet to the up layers
> 
> And ehea_check_cqe() checks for wrong checksumed packets on:
> 	
>          if ((cqe->status & EHEA_CQE_STAT_ERR_MASK) == 0)
>                  return 0;
> 
> 
> Botton line, TCP/UDP packets with wrong checksums are dropped by 
> ehea_proc_rwqes(), others go to the up layer.
> 
> So, back to your question, you are saying that we shouldn't do that, 
> meaning that we should send to the upper layers all packets ? even those 
> that have the wrong checksum ?
> 

I am pretty sure most (if not all) netdev drivers pass the packet with
invalid checksum to upper stack, so that we can increment appropriate
SNMP counters, in IP stack or UDP/TCP/whatever stack.

tg3, bnx2, e1000, skge, sky2, bnx2x, niu, r8169, igb, ... seems to do
that.