Re: [OSPF] OSPFv3 Extended LSAs TLV-level "disposition-if-unsupporetd indicator"?

[David Lamparter <[e-mail]>]

On Tue, Aug 06, 2013 at 02:31:07PM +0000, Acee Lindem wrote:
> On Aug 6, 2013, at 10:21 AM, David Lamparter wrote:
> > On Tue, Aug 06, 2013 at 02:16:10PM +0000, Acee Lindem wrote:
> >> I don't think this is a good idea at all. An implementation can only
> >> deal with the TLVs it understands so the ignore options don't make
> >> sense.
> >> If the other actions are necessary, we'd be better served with a more
> >> definitive use case and encoding than reusing TLV bits.
> >
> > Section 3 of the draft states:
> > 	[...] Unrecognized types are ignored.
> > How does that make sense then?  Did I misunderstand that sentence?
>
> No - the only valid option for a TLV that you don't recognize is to
> ignore it. It makes no sense to attempt to imply some action based on
> bits in the TLV type.

Hm.  We have the same thing with U/S1/S2 bits in LSA function codes.
Other protocols have a "critical" bit indicating ignore/error behaviour.
I've tried to adapt that for a routing use case - I guess I'm not seeing
the problem?  Do you have a few minutes to explain in more depth?

(I guess it needs a little more thinking in particular regarding
multiple TLVs with different unsupported disposition, probably just an
ordering and "choose worst".  I've also implied without saying that all
LSAs are flooded unmodified, i.e. U=1.)


-David

> >> On Aug 6, 2013, at 9:49 AM, David Lamparter wrote:
> >>> 00 - ignore TLV
> >>> this can be used for all "hint" TLVs, stuff like maybe communicating
> >>> the origin ASN for external routes or whatever you can dream up.
> >>> 
> >>> 01 - ignore parent
> >>> on calculating SPF, completely ignore the resulting route.  This is
> >>> useful for MT-OSPF (if it ever happens), to be used on a MT-ID TLV
> >>> with an MT-ID != 0.  Basically, non-MT routers can ignore all nonzero
> >>> MT topologies this way.
> >>> 
> >>> 10 - strong unreachable
> >>> mark the route's destination prefix as unreachable and install a
> >>> corresponding blackhole/... route.  This is the right thing to do on
> >>> SADR routes when they hit a non-SADR router.  Even if we have the same
> >>> prefix reachable on a non-SADR route with a lower metric, we can't
> >>> ensure that it's loop-free for a particular source address.
> >>> 
> >>> 11 - weak unreachable (?)
> >>> treat the route as "unreachable", adding it to the SPF result as such,
> >>> if there is no shorter path to the same prefix so far.  This is
> >>> probably the least useful type, I can only come up with something like
> >>> "route that requires special encapsulation (tunnel?)" - no idea on the
> >>> reality here.
>