The MPLS WG Archive

[David Charlap <David.Charlap@marconi.com>]

David Charlap wrote:
> Sandeep B wrote:
>>
>> If all vendors support MPLS as it was intended, only egress and
>> penultimate hop should be checking for L3PID. No intermediate LSR
>> should be even noticing what is inside the label.
> 
> Absolutely correct.  A transit router shouldn't care at all what's
> inside the packets.  Only nodes that pop labels (and really only
> those that pop the last label off the stack) need to know about the
> L3PID, since they are the only nodes who will be responsible for
> forming an appropriate layer-2 header.
> 
>> Few of the biggest names in industry do check on L3PID at every
>> LSR and reject if its not IP.

I should just add that this isn't necessarily unreasonable.

Due to the nature of RSVP, a router does not know whether or not it will
have to pop a label during Path processing.  Yes, it will know if it is
the egress node, but if it's not, it can't know in advance if it will be
given a label to swap to or a request to pop the label (implicit NULL).

Also note that when the route table changes, a pure transit router may
become a penultimate-hop router.

In other words, a router can't know for certain that it will never be
asked to pop a label stack.  The result is that it may make sense to
validate the L3PID at the time of Path processing.

It could wait for the Resv message, and only validate L3PID if it has to
do a pop, but that may not be efficient for all implementations.  It
would also increase the overhead of cleanup after generating the error.

But none of this has anythng to do with the fact that RSVP's L3PID must
be an Ethernet protcol ID.

-- David