Re: [PATCH net-next V4 03/13] bridge: Validate that vlan is permitted on ingress

[Shmulik Ladkani <shmulik.ladkani@gmail.com>]

On Thu, 20 Dec 2012 10:41:28 -0500 Vlad Yasevich <vyasevic@redhat.com> wrote:
> >> +static bool br_allowed_ingress(struct net_bridge_port *p, struct sk_buff *skb)
> >> +{
> >> +	struct net_port_vlan *pve;
> >> +	u16 vid;
> >> +
> >> +	/* If there are no vlan in the permitted list, all packets are
> >> +	 * permitted.
> >> +	 */
> >> +	if (list_empty(&p->vlan_list))
> >> +		return true;
> >
> > I assumed the default policy would be Drop in such case, otherwise
> > leaking between vlan domains is possible.
> > Or maybe, ingress policy when port isn't a member of ingress VID should
> > be configurable (drop/allow).
> 
> We have have to default to allow since we want to retain original bridge 
> functionality if there is no configuration.

Ok; so having the port not a member of ANY vlan is a "port vlan
disabled" configuration knob, and as such, it is a member of ANY vlan,
meaning that:

(1) every "non-vlan port" is connected to any other "non-vlan port"
(2) frame ingress on a "non-vlan" port may egress on a "vlan enabled"
  port, depending on the ingress VID and the port-membership map of the
  egress port
  (and thus, PVID should be defined even to "non-vlan" ports, for the
  case where untagged frame is received on the non-vlan port)
(3) frame ingress on a "vlan-enabled" port would always egress on
  "non-vlan" ports

Seems ok.
However this is an additional nuance that might not be expected by the
user configuring the bridge; maybe this needs some clarification.

> >> +	vid = br_get_vlan(skb);
> >> +	pve = nbp_vlan_find(p, vid);
> >
> > Why search by iterating through NBP's vlan_list?
> > You know the VID (hence may fetch the net_bridge_vlan from the hash), so
> > why don't you directly consult the net_bridge_vlan's port_bitmap?
> 
> It's an alternative...  I am betting that this port isn't in too many 
> vlans and that searching the list might be faster.

I assumed the opposite: finding the hash bucket is just a bitwise mask,
and number of items in a bucket would rarely be grater than 1.
I expect such code to be shorter, but this needs to be verified.

Regards,
Shmulik