RE: [IPV6] Q: corrupt checksums when transferring data

Stephen Hemminger wrote:

On Fri, 25 Aug 2006 11:13:48 -0700
"Brandeburg, Jesse" [off-list ref] wrote:

quoted

I'm enabling e1000 to offload IPv6 since the 2.6.18+ kernels support
it. The kernel I'm testing is 2.6.18-rc4.

Yes, something is wrong with the GSO code. I am bisecting this bug
	http://bugzilla.kernel.org/show_bug.cgi?id=7050


It looks like GSO is handing an IPV6 segment down to the sky2 driver
even though it asks for only NETIF_F_TSO.

Ah ha, I was wondering if that bug report on sky2 might be related to
this issue.  E1000 actually sends the data I think (it just has a bad
checksum) when handed a too long frame.  Seems like the stack should
never give us something longer than the MTU + enet header, esp with all
hardware offloads disabled.

So I have a very easy repro with netperf
on remote: netserver -4 -6
netperf -H lh2-6,6 -t TCP_MAERTS -- -m4K -S128K -s128K

The remote will generate the bad frames.

Jesse