On Thu, Nov 15, 2012 at 9:02 AM, Ansis Atteka <aatt...@nicira.com> wrote:
> On Wed, Nov 14, 2012 at 1:48 AM, Jesse Gross <je...@nicira.com> wrote: > > On Tue, Nov 13, 2012 at 9:52 AM, Ansis Atteka <aatt...@nicira.com> > wrote: > >> > >> diff --git a/datapath/actions.c b/datapath/actions.c > >> index 76c9823..5da4b74 100644 > >> --- a/datapath/actions.c > >> +++ b/datapath/actions.c > >> @@ -435,6 +435,10 @@ static int execute_set_action(struct sk_buff *skb, > >> skb->priority = nla_get_u32(nested_attr); > >> break; > >> > >> + case OVS_KEY_ATTR_SKB_MARK: > >> + skb->mark = nla_get_u32(nested_attr); > > > > > > One other thing that I forgot - skb->mark used to be called nfmark > (before > > 2.6.20) and in that case only existed if CONFIG_NETFILTER was defined so > we > > should handle that case as well. If it doesn't exist then on the match > we > > can treat it as zero and if someone tries to use an action then we should > > deny it on flow setup. > > Currently lowest supported kernel version is 2.6.18. Perhaps it is > worth to bump it up to 2.6.20? > There aren't any distributions that I know of that use 2.6.20, so for practical purposes that would really move us up to 2.6.26. It might be fine at this point (2.6.18 is more than 6 years old) but the compatibility code for this is so minor that it doesn't seem worth dropping support over it. > I guess this skb mark get/set compatibility code should go into > skbuff.h, right? I would put it in compat.h since it doesn't correspond to a current or proposed upstream kernel function. > With "denying flow on setup" do you mean to return > -EINVAL in validate_set() (if there is skb mark set action, but kernel > does not support netfilter marks)? > Yes, that's right.
_______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
