jon.otterholm at ide.resurscentrum.se
Mon Nov 21 04:45:48 PST 2005
You got it all right.
Antispoof sounds nice.
The reason why I have to proxy-arp mac between VLANs is that one mac
cannot end up mapped to more than one port in the switches FDB. If they
do - we get something called "host-flapping" on IOS-language.
On Mon, 2005-11-21 at 11:28 +0000, Brian Candler wrote:
> > > On Thu, Nov 17, 2005 at 04:52:03PM +0100, Jon Otterholm wrote:
> > > > Scenario#1:
> > > > -I have a range of ip's, for example 126.96.36.199 - 188.8.131.52.
> > > > -I want to distrubute theese ip's to my customers via DHCP.
> > > > -They are all atached to me via a VLAN-trunk on a unique VID
> > > > -I have 200+ customers.
> Let me see if I can summarise your requirements:
> cust1 --------,
> \ service network
> vlan2 \ trunked single subnet x.y.z.0/24
> cust2 -----------===============*------+------------+-------
> / | |
> vlan3 / DHCP srv router ---> Internet
> cust3 --------'
> The constraints are:
> - cust1, cust2, cust3 are all on the same subnet x.y.z.0/24 and get an IP
> address allocated by DHCP
> - However, the MAC address for cust1 must never appear as a source MAC
> address on vlan2 or vlan3, as this would confuse the provider's
> trunked switching infrastructure
> So you can't just bridge all the VLANs together. Rather:
> 1. a broadcast from custX must appear on the service network, but not
> on any of the other customer VLANs. More strongly, no packet from
> custX may appear on any other VLAN apart from the service network. [*]
> 2. a broadcast from the service subnet should appear on all customer VLANs
> (e.g. ARP from DHCP server or router)
> 3. an ARP request for custY from custX must be proxy-ARP responded to by
> a device on the service network, otherwise customers wouldn't be able to
> communicate with each other.
> 4. a unicast packet from the service network to a customer must be forwarded
> to the correct customer VLAN
> Is that a reasonable summary? Taking that as the base:
> Point (4) implies that a bridge forwarding table must still be built,
> because the device performing this function (labelled '*' in the above
> diagram) needs to associate a MAC address with a VLAN, and do so by
> learning rather than static configuration.
> Point (1) says that you can't just bridge all the VLANs together, because
> otherwise a packet to a broadcast address or to an unknown MAC address would
> be forwarded to all the other VLANs. We want this to happen for packets
> originating from the service network (point (2)), but not for packets which
> originate from the customer networks. So, some sort of L2 forwarding filter
> should do the trick: configure it so that packets may only be forwarded to
> customer VLANs if they originate from the service network. If this filter is
> applied, you guarantee that a customer's MAC address will never appear as a
> source on any other VLAN.
> Point (3), proxy ARP, is easy enough. You know all the possible customer IPs
> - they are exactly the range assigned to the DHCP server to allocate - and
> therefore you can proxy-ARP respond for any IP address within the DHCP
> range, as long as the request originated from a customer VLAN (which can
> also be determined by the ARP source IP). The router itself could perform
> this proxy ARP function, or else any server on the service network running
> something like choparp (which can give out the router's MAC address in the
> ARP responses). IP datagrams from custX to custY then go
> So actually, when thought about like this, the L2 masquerading requirement
> vanishes, and what you really need is bridging plus some L2 filtering based
> on ingress and egress interfaces.
> Unfortunately, I don't know if FreeBSD has this level of L2 filtering (I
> note that the bridge(4) documentation says that ipf/ipfw filtering only
> works for IP datagrams). However a frob on the bridging code should be
> possible; call the first interface 'master' and the rest 'slaves', and have
> a rule so that packets to a 'slave' interface are only forwarded if they
> originated from the 'master' interface.
> Aside: the network as designed above has an obvious flaw that any customer
> can DHCP for as many different machines as they wish, and therefore exhaust
> your DHCP pool. You could have a separate mini DHCP server listening on each
> VLAN and only handing out a single IP, but that doesn't stop customers
> stealing other customer's IPs through static configuration. So actually, I
> think you need anti-spoofing filters on each VLAN too.
> Doing that, you end up statically routing a separate IP down each VLAN, in
> which case what you *really* want is to be able to configure each VLAN
> subinterface as if it were a point-to-point interface. But I don't think
> FreeBSD supports that on broadcast media.
> [*] Or if it did appear on the other customer VLANs, it would have to be
> masqueraded to appear as if it came from a MAC address on the service
> network; however I believe this isn't actually necessary, as the only
> broadcasts we really care about here are ARP requests. All others
> can be dropped, and indeed probably should be dropped so that all
> your customers don't get drowned in each other's broadcast traffic.
More information about the freebsd-net