Well, *somebody* needs to advertise the address used as the source
address for OSPF on unnumbered links, or the protocol will fail. The
spec specifically says that the receiver of said packets is supposed
to advertise that address. The spec does *not* say that the owner of
that address should advertise it, as stupid as that may seem.
Speaking of intent is all well and good, but the spec has been out
there for about 20 years now and it says something else. John Moy's
implementation is not the spec.
In reality, implementors seem to end up doing the right thing anyhow.
(I still get grief for advertising the loopback address by default in
our implementation, 11 years on, as that's not in the spec either.)
But the spec, as written, works. Simply ignoring Option 1 for
unnumbered links *breaks the spec*. Unless the spec is changed to
specifically require that the address used as the source on unnumbered
links is advertised in *some* LSA by the owner of said address, we
cannot say that the clause as written is in error.
--Dave
On Jan 6, 2009, at 6:41 PM, Acee Lindem wrote:
Hi Dave, Joakim,
The intent of the protocol is to not advertise stub links in router
LSAs for unnumbered links. If one wishes to advertise the local
address an unnumbered link is using, the interface should be
configured as an OSPF interface. Unfortunately, RFC 2328 section
12.4.1.1 doesn't state this explicitly. Although I wasn't involved
in the generation or review of RFC 2328, John Moy's public domain
implementation (available at www.ospf.org) confirms this behavior
and is consistent with most implementations - though possibly not
all implementations.
Thanks,
Acee
On Jan 6, 2009, at 1:46 PM, Dave Katz wrote:
The idea of Option 1 is to provide connectivity to the remote
router itself. The remote router has an address, and it is known
to the local router as soon as the first Hello is received. So by
definition it is always possible to do (since the remote router's
address must be known in order to have an adjacency in the first
place.)
One could claim that it would make more sense for a router to
announce its *own* address in a link, and in fact some
implementations do so gratuitously for safety's sake. Or a
loopback interface (with an independent address) is included in the
OSPF configuration to achieve the same thing. But without either
Option 1 or one of the hacks, it may not be possible to address the
router itself for management purposes.
So I would not suggest to omit the second link; the inaccuracy is
in the example instead.
--Dave
_______________________________________________
OSPF mailing list
OSPF@ietf.org
https://www.ietf.org/mailman/listinfo/ospf
