I have a typical router with multiple networks on
192.168.1.0/25
192.168.1.128/26
192.168.1.192/28
192.168.1.255/32
etc.

This leads to a dozen or so routes advertised by OSPF, and this makes a messy routing table when there are 5 or 6 of these routers connected together.

If I add a static route for the whole /24 to the loopback interface, using "ip route add dst-address=192.168.1.0/24 gateway=bridge-loopback", and use redistribute-static in OSPF to distribute, this allows me to filter at the OSPF/BGP boundry using routing filter to only distribute prefix length of /24 and nothing smaller.

However I would ideally like just the /24 to be advertised on OSPF too. In theory using an ospf-out routing filter would work, but doesn't seem to apply to directly connected networks, which I suspect is related to "and internal (intra-area) OSPF filtering is not supported yet" in the manual.

I can filter these out on the ospf-in, but is there another way to effectively apply ospf-out to the directly connected routes - or another way to do what I want (only advertise the /24 by OSPF)

It is btter advertise connected networks as it is and summarise them on abr.

As Anumrak suggests, OSPF ABR (area boundary routers) were designed to do precisely this.

Configure the upstream router as follows: Configure the router at SiteA to have an Area ID of '1' (aka 0.0.0.1). This will result in SiteA receiving a default gateway from the ABR and anything within 192.168.1.0/24 being summarised.

The entire network looks like

With Box 1 / 2 / 3 being able to plug into either Edge 1 or Edge 2 depending on the day (but not into both locations at the same time). Edge 1 or Edge 2 then advertises the box into the core.

Edge1 advertises via BGP:
* 192.168.1.0/24
* 192.168.11.0/24
* 192.168.12.0/24

A route filter stops 192.168.0.0/24 from being advertises by BGP (otherwise the network would be advertised from both Edge1 and Edge2, and that's wrong)

If Box2 is removed, Edge1 no longer advertises 192.168.12.0/24. If Box2 is plugged into Edge2, then Edge2 advertises 192.168.12.0/24 instead. (Box2 can't be plugged into both as they are miles away from each other)

As OSPF is p2mp, Edge1 also advertises 192.168.0.11/32 and 192.168.0.12/32, which is good for traceroute and pingability. However Edge1's address (192.168.0.1/32) isn't advertised itself (and I block the 192.168.0.0/24 network from being advertised)

Each of the OSPF areas is independent, so in any system there's just one - area 0. I use BGP between the Core and the Edge, so multi areas and summarisation doesn't really work.

Would I be right in thinking that would mean a private area for each box - that never leaves that box - containing the local network, and summarising into the area - making each box/edge an ABR in itself?

One more question: to have 192.168.0.1/32 advertised into BGP from Edge1, I'm unsure which is best
* Add a second /ip address line to the OSPF interface
ip address add address=192.168.0.1/24 interface=ether_ospf
ip address add address=192.168.0.1/32 interface=ether_ospf

* Add a blackhole route and redistrubute statics into bgp
ip route add dst-address=192.168.0.1/32 type=blackhole

* Add a static route to the OSPF interface and redistrubute statics into bgp
ip route add dst-address=192.168.0.1/32 gateway=ether_ospf

I prefer originating BGP prefixes via static routes as it provides a single control point to:

• Apply BGP communities
• Blackhole traffic to suppress ping pongs
• Avoid having to manage BGP 'networks'

Sample: You can then write routing filter rules to:

• Accept and blackhole on iBGP sessions (when community contains 64512:900)
• Set outgoing MED, prepend or apply upstream BGP community to manipulate local preference (64512:4000 could designate Router 1, 64512:4010 Router 2, etc)
• Accept outgoing advertisements on eBGP sessions (filtering on 64512:500 for your own subnets, you could mark down steam customers with another and peers individually or again by location (preferably both))

I do not redistribute anything in to OSPF, purely defining networks covering loopbacks and point-to-point links. MPLS LDP and OSPF timers are reduced to 1s hellos and 10s hold time. iBGP peering sessions have 'nexthop-choice=force-self update-source=lo' set.
We accept primarily a default route from our upstream BGP provider and set OSPF to 'distribute-default=if-installed-as-type-1'.
We peer each provider edge router with redundant route reflectors, but you don't need to concern yourself with that if you have 2 routers, as they will be meshed. BGP instances have 'redistribute-connected=yes redistribute-static=yes' set.

PS: The following route filter can be used to selectively avoid redistribution to BGP (simply add 'origin=egp' to selective static routes):