The MPLS WG Archive

[Bob Thomas <rhthomas@cisco.com>]

Manoj,

> While going through RFC 3036 (LDP) section A.1.4 (label release
> processing ), I came across the following section :
> 
> 
> LRl.5     LRl.5  Has LSR previously advertised a label for this FEC  to
> other peers ?
>               If so, goto LRl.10
> 
> If this check fails (ie LSR has not advertised any labels to other peers
> for this FEC), the corresponding downstream label may be released. Now,
> consider the case below :
> 
> There are three LSRs        Ra-----Rb-------Rc connected in a striaght
> line as shown (Rc being the downstream LSR for Rb and so on).
> Now lets assume Ra is non merge capable and Rb is merge capable.
> For a given FEC F, Ra requests multiple labels from Rb (one for each of
> Ra's upstream peers for F). In response to these requests, Rb gives
> label mappings {Lb1, Lb2, Lb3.....} to Ra and binds all these mappings
> to Label Lc received from Rc for F. Now Ra releases one of the labels,
> say Lb1, which it got from Rb for F. Now according to LRl.5,  Rb checks
> if there are any other peer holding labels for Lc (downstream label from
> Rc for FEC F). The check fails as Ra is the only peer holding labels for
> F. In this case, Rb may release the downstream label Lc (if release
> propagation is configured in Rb). This leaves other labels like {Lb2,
> Lb3...}
> sent to Ra from Rb (for FEC F) without any outgoing label on Rb. Thus
> the packets from Ra labelled with
> {Lb2, Lb3 ...} will need to use ip forwarding to reach Rc.
> 
> I think we should not release a downstream label as long as there is
> even one upstream label mapped to it but  the procedure described
> doesn't seem to take care of this case. Let me know if there is anything
> I am missing here.

I think you have found an error in LRl.5.  The intent is as you
suggest, but LRl.5 doesn't correctly specify that intent.  LRl.5 should
be saying something like:

   LRl.5   Has LSR previously advertised a label for this FEC to other
           peers or another label for this FEC to MsgSource?
           If so, goto LRl.10.

or:

   LRl.5   Does LSR have outstanding label advertisements for this FEC?
           If so, goto LRl.10.


Thanks,
Bob