Re: TCP checksum offloading (complete)

Casey Carter <Casey@Carter.net> · Thu, 13 Nov 2003 07:41:51 -0600

The intuition here is to think about exactly what needs to be subtracted 
from the whole-packet checksum to get the TCP checksum: the IP header 
checksum.  And the IP header checksum, since we know that it must be 
correct, must be zero.  Subtracting zero from a checksum leaves it 
unchanged.  Thus, the subtraction is implicit.

Nick Patavalis wrote:

On Thu, Nov 13, 2003 at 12:21:35PM +0100, francois donzet wrote:

But by having a look at the tcp_v4_recv() function,
and more precisely inside tcp_v4_checksum_init() code,
skb->csum is used 'as is' when skb->ip_summed is set
to CHECKSUM_HW. There is no substraction at all.
That's the problem :/

That was my conclusion too, after browsing through the kernel code. Do
we miss something, or does the "substraction" of the extra fields is a
responsibility of the driver?

/npat

--
Casey Carter
Casey@Carter.net
ccarter@cs.uiuc.edu
AIM: cartec69

-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html