The MPLS WG Archive

[John.Brennen@marconi.com]

I think I have found a serious protocol problem in LDP (RFC 3036).
The best way to describe it is to give an example...


Consider the following set of events, in chronological order
 (all messages refer to the same FEC/Label pairing; the session
 supports unsolicited label mappings; and Router B is using
 conservative label retention):


  Router A                          Router B
  --------                          --------

  Sends Label Mapping
                                    Receives Label Mapping
                                    Sends Label Release
                                     (because Router A is not the next hop)
  Sends Label Mapping
   (because the hop count changed)
                                    Router A becomes the next hop
  Receives Label Release
                                    Receives Label Mapping


The problem here is that the Label Mapping message which updates the
hop count and the Label Release message are sent more or less
simultaneously.

The result in the scenario above is that Router A thinks that there is no
valid Label Mapping advertised to Router B.  Router B thinks that it
has a valid Label Mapping.

At least three problems can come from this.  First, Router B can put traffic
onto a label which Router A has reused for another purpose.

Second, Router A may try to readvertise the FEC in the future.
Router B will not accept any such Label Mappings (LMp.10, Appendix A,
RFC 3036), unless Router A is fortunate enough to use the same
label which was originally used.

Third, if Router B tries to release the label, Router A may
report an error.


Is this analysis correct?  Is this a real problem with the protocol?
If so, is it already known?  How can this problem be solved?

    Jack Brennen