The MPLS WG Archive

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

[RE: ICMP generation on TTL expiry]

Eric Gray wrote:
> 
>     Hmmm.  Thought it might be that.  The answer is that
> you would not be likely to be able to send an ICMP error
> message to the packet source if you would not be able to
> route the original packet - if unexpired - to the packet
> destination.

You may have a route to the source, even if you don't have one
to the destination.

>     However, there is a back-doorsy kind of way to send
> an ICMP error message to the LSP ingress if the LSP was
> established using either CR-LDP or RSVP-TE.  Also, I
> am pretty sure that Ron Bonica's efforts to resolve this
> little problem in the general tunneling case have not yet
> been abandoned.
> 
>     And I sure don't see how having an L3 PID identifying
> an LSP as having been established to carry non-IP packets
> is going to help with this specific problem (therefore the
> question on relevancy).

If the packet is not IP, then the source won't expect or want an ICMP
response.  It might, however, expect something else, depending on what
that L3 protocol is.

If the L3 protocol is unknown, there isn't much it can do.  If it knows
what the protocol is (IP or otherwise), it can generate an appropriate
response (assuming, of course, that the response can be routed back to
the source.)

The mechanism with which it uses to know the protocol doesn't matter, as
long as it knows.

-- David