Thank you for remind plain text.
Use csum_start to seperate these two cases, maybe a good idea.
1.Disable tx csum
skb->ip_summed == CHECKSUM_PARTIAL && skb_transport_header == udp
2.Enable tx csum
skb->ip_summed == CHECKSUM_PARTIAL && skb_transport_header != udp
Correct?
在 2023/4/3 下午6:56, Edward Cree 写道:
On 02/04/2023 19:18, Willem de Bruijn wrote:
On Fri, Mar 31, 2023 at 10:31 PM Fei Cheng <chenwei.0515@xxxxxxxxxxxxx> wrote:
From: "chenwei.0515" <chenwei.0515@xxxxxxxxxxxxx>
If vxlan-dev enable tx csum offload, there are two case of CHECKSUM_PARTIAL,
but udp->check donot have the both meanings.
1. vxlan-dev disable tx csum offload, udp->check is just pseudo hdr.
2. vxlan-dev enable tx csum offload, udp->check is pseudo hdr and
csum from outter l4 to innner l4.
Unfortunately if there is a nat process after vxlan tx，udp_manip_pkt just use
CSUM_PARTIAL to re csum PKT, which is just right on vxlan tx csum disable offload.
In case 1 csum_start should point to the (outer) UDP header, whereas in
case 2 csum_start should point to the inner L4 header (because in the
normal TX path w/o NAT, nothing else will ever need to touch the outer
csum after this point).
The issue is that for encapsulated traffic with local checksum offload,
netfilter incorrectly recomputes the outer UDP checksum as if it is an
unencapsulated CHECKSUM_PARTIAL packet, correct?
So if netfilter sees a packet with CHECKSUM_PARTIAL whose csum_start
doesn't point to the header nf NAT is editing, that's exactly the case
where it needs to use lco_csum to calculate the new outer sum. No?
-ed
PS. Fei, your emails aren't reaching the netdev mailing list, probably
because you're sending as HTML. Please switch to plain text.
