On Wed, Jul 24, 2024 at 3:56 PM Ilya Maximets <[email protected]> wrote:
>
> On 7/12/24 23:30, Mike Pattrick wrote:
> > When prepending a tunnel header to a packet marked for segmentation, we
> > need to adjust the segment size. Failure to do so can result in packets
> > that are larger than the intended MTU post segmentation.
>
> Hi, Mike. Not a full review yet, but could you, please, explain in
> more details why we need to adjust TSO segment size here?
>
> In order to ensure packet delivery, MTU inside VM / container must be
> smaller than the physical network MTU by the size of the tunnel header.
> So, tso_segsz of the packet received from such a VM or container should
> be already small enough. Or am I missing something?
My rationale here is the VM can't know if OVS will be encapsulating a
packet in a udp tunnel, or how big that encapsulation would be. The
UDP encapsulation is pretty large and potentially variable, it may be
overly burdensome if the VM MTU must be adjusted to anticipate any UDP
tunnel.
Thanks,
Mike
>
> Best regards, Ilya Maximets.
>
> >
> > Fixes: 084c8087292c ("userspace: Support VXLAN and GENEVE TSO.")
> > Signed-off-by: Mike Pattrick <[email protected]>
> > ---
> > lib/netdev-native-tnl.c | 11 +++++++++++
> > 1 file changed, 11 insertions(+)
> >
> > diff --git a/lib/netdev-native-tnl.c b/lib/netdev-native-tnl.c
> > index 16c56608d..78c91b5fa 100644
> > --- a/lib/netdev-native-tnl.c
> > +++ b/lib/netdev-native-tnl.c
> > @@ -161,6 +161,17 @@ netdev_tnl_push_ip_header(struct dp_packet *packet,
> > const void *header,
> > *ip_tot_size = dp_packet_size(packet) - sizeof (struct eth_header);
> >
> > memcpy(eth, header, size);
> > +
> > + /* The prepended header may cause TSO marked packets to exceed the
> > intended
> > + * MTU on segmentation. */
> > + if (dp_packet_hwol_is_tso(packet)) {
> > + uint16_t tso_segsz = dp_packet_get_tso_segsz(packet);
> > + if (tso_segsz > size) {
> > + tso_segsz -= size;
> > + dp_packet_set_tso_segsz(packet, tso_segsz);
> > + }
> > + }
> > +
> > /* The encapsulated packet has type Ethernet. Adjust dp_packet. */
> > packet->packet_type = htonl(PT_ETH);
> > dp_packet_reset_offsets(packet);
>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
