tracing a packet through the kernel stack

Hi there, 
i am trying to trace a packet which checksum has been
calculated in hardware (skb->ip_summed set to
CHECKSUM_HW) when it comes to the tcp layer.

I start with net/tcp_ipv4.c/tcp_v4_recv(). Here, there
is a call to tcp_v4_checksum_init(). If the call
returns a negative number, the packet is considered
invalid (checksum problem), and we go to the label
bad_packet.

i looked at tcp_v4_checksum_init() (in tcp_ipv4 too).

When skb->ip_summed  is set to CHECKSUM_HW, we first
set it to CHECKSUM_UNECESSARY, to indicate that
checksum hasn't to be computed later. Then we call
tcp_v4_check() function, with params including the
skb->len value, the source ip address,destination ip
address, and the value of skb->csum.

As far i know, skb->csum contains a sum of the entire
packet contents, computed by the hardware which
received the packet at the link layer level.

To sump up, We have :

tcp_v4_recv()-->tcp_v4_checksum_init()-->tcp_v4_check()-->csum_tcpudpmagic().

When i look at the architecture dependent code
(asm-x86 for me) :

csum_tcpudp_magic is followed by csum_tcpudp_nofold,
and csum_fold calls.

In my understanding of the checksumming operations,
csum_tcpudp_magic() computes the pseudo header
checksum, based on the source adress, the destination
adress, the length of the packet and the ip protocol
of the packet (here tcp). 

The initial sum to be add in contains the value of
skb->csum. My initial believing would have been that
(as someone suggests me after), some substractions
from the skb->csum (as pseudoheader contains fields in
ip header) would be made before checking the entire
TCP checksum.

But here, the skb->csum value is used "as is" in the
TCP checksum computation.

If you can give a little help, or tell me if i am
looking the wrong part of code, thanks.

___________________________________________________________
Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
Yahoo! Mail : http://fr.mail.yahoo.com