Re: [PATCH] openvswitch: Orphan frags before sending to userspace via Netlink to avoid guest stall

[Thomas Graf <tgraf@redhat.com>]

On 02/28/2014 08:16 PM, Zoltan Kiss wrote:
> The kernel datapath now switched to zerocopy Netlink messages, but that also
> means that the pages on frags array are sent straight to userspace. If those
> pages came outside the kernel, we have to swap them out with local copies.
>
> Signed-off-by: Zoltan Kiss <zoltan.kiss@citrix.com>
> ---
>   net/openvswitch/datapath.c |    6 ++++++
>   1 file changed, 6 insertions(+)
>
> diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
> index 36f8872..ffb563c 100644
> --- a/net/openvswitch/datapath.c
> +++ b/net/openvswitch/datapath.c
> @@ -464,6 +464,12 @@ static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
>   	}
>   	nla->nla_len = nla_attr_size(skb->len);
>
> +	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) {
> +		err = -ENOMEM;
> +		skb_tx_error(skb);
> +		goto out;
> +	}
> +
>   	skb_zerocopy(user_skb, skb, skb->len, hlen);
>
>   	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */

How about we move the call to skb_orphan_frags() into skb_zerocopy()
itself and call it before we actually reference the frags?

It's not just OVS affected but nfqueue as well.