Re: [PATCH net-next] skbuff: clear tx zero-copy flag

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Jul 25, 2011 at 04:40:57PM +0800, Herbert Xu wrote:
> On Mon, Jul 25, 2011 at 11:07:43AM +0300, Michael S. Tsirkin wrote:
> >
> > However macvtap passes an skb directly to the
> > lower device, so as long as macvtap is the only user
> > of that interface, we are fine I think - there's
> > no way for an skb to get from macvtap to splice
> > read path I think.
> > 
> > Right?
> 
> Yes, as long as you can guarantee that the skb never loops back
> then you should be fine.
> 
> However, does macvtap really bypass everything, including the
> qdisc layer? The qdisc layer is certainly capable of looping
> the skb back with the redirect action.
> 
> Cheers,

No, I don't think macvtap bypasses the qdisc.
Is the action in question here?
static int tcf_mirred(struct sk_buff *skb,
			const struct tc_action *a,
                      struct tcf_result *res)

if yes that seems to always clone an skb, which in turn
does the copy so we are fine?

> -- 
> Email: Herbert Xu <herbert@gondor.apana.org.au>
> Home Page: http://gondor.apana.org.au/~herbert/
> PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt