On 6/5/25 21:48, Jakub Kicinski wrote:
On Thu, 5 Jun 2025 21:33:26 +0700 Bui Quang Minh wrote:
On 6/5/25 18:03, Paolo Abeni wrote:
On 6/3/25 5:06 PM, Bui Quang Minh wrote:
In virtio-net, we have not yet supported multi-buffer XDP packet in
zerocopy mode when there is a binding XDP program. However, in that
case, when receiving multi-buffer XDP packet, we skip the XDP program
and return XDP_PASS. As a result, the packet is passed to normal network
stack which is an incorrect behavior.
Why? AFAICS the multi-buffer mode depends on features negotiation, which
is not controlled by the VM user.
Let's suppose the user wants to attach an XDP program to do some per
packet stats accounting. That suddenly would cause drop packets
depending on conditions not controlled by the (guest) user. It looks
wrong to me.
But currently, if a multi-buffer packet arrives, it will not go through
XDP program so it doesn't increase the stats but still goes to network
stack. So I think it's not a correct behavior.
Sounds fair, but at a glance the normal XDP path seems to be trying to
linearize the frame. Can we not try to flatten the frame here?
If it's simply to long for the chunk size that's a frame length error,
right?
Here we are in the zerocopy path, so the buffers for the frame to fill
in are allocated from XDP socket's umem. And if the frame spans across
multiple buffers then the total frame size is larger than the chunk
size. Furthermore, we are in the zerocopy so we cannot linearize by
allocating a large enough buffer to cover the whole frame then copy the
frame data to it. That's not zerocopy anymore. Also, XDP socket zerocopy
receive has assumption that the packet it receives must from the umem
pool. AFAIK, the generic XDP path is for copy mode only.
Thanks,
Quang Minh.
