ACK: [PATCH 0/3][Yakkety][SRU] virtio: fix xmit packet flag settings
Seth Forshee
seth.forshee at canonical.com
Thu Apr 20 13:56:49 UTC 2017
On Wed, Apr 19, 2017 at 04:41:44PM -0700, Jay Vosburgh wrote:
> BugLink: https://bugs.launchpad.net/bugs/1683947
>
> Beginning with fd2a0437dc33 ("virtio_net: introduce
> virtio_net_hdr_{from,to}_skb"), virtio_net incorrectly sets
> VIRTIO_NET_HDR_F_DATA_VALID on CHECKSUM_UNNECESSARY packets. If these
> packets are modified prior to transmission (e.g., by NAT), they are
> dropped by the receving virtio. I suspect (but have not confirmed
> experimentally) that this is because the checksum is incorrect after NAT
> modifies the packet, but the checksum is not recomputed.
>
> Patch 1/3, 3e9e40e74753 ("virtio_net: Simplify call sites for
> virtio_net_hdr_{from, to}_skb()."), is a prerequisite to permit patch 3/3,
> 6391a4481ba0, to apply more easily and end up with the final code closer
> to mainline.
>
> Patch 2/3, 501db511397f ("virtio: don't set
> VIRTIO_NET_HDR_F_DATA_VALID on xmit") was found by bisect to initially
> resolve the issue.
>
> Patch 3/3, 6391a4481ba0 ("virtio-net: restore
> VIRTIO_HDR_F_DATA_VALID on receiving") is a follow-on to 501db511397f for
> the complete fix.
>
> As the issue was introduced during 4.8-rc and resolved before 4.10
> release, this bug exists only in the Ubuntu Yakkety 4.8 kernel series.
Patches are clean cherry picks with positive tests results. Assuming the
upstream sha-1's are fixed up as Colin pointed out:
Acked-by: Seth Forshee <seth.forshee at canonical.com>
More information about the kernel-team
mailing list