Vince, Dino,

I wrote:

I'd still like to understand the ICMP message situation better. I realize
that the traffic is supposed to be unidirectional, but how would, say,
lower MTU somewhere along the path be dealt with? I'm not necessarily
asking for a change in what the spec does, but it should be clear about the
consequences.

Vince wrote:

Under normal circumstances, the only thing that should be send across the
ALT would be a Map Request, which should not cause MTU problems. Of course,
if a (deprecated) Data Probe were used, then all bets are off.

If a Map Request or Data Probe were to hit an MTU problem or otherwise be
undeliverable across the ALT and no ICMP message could be returned from the
ALT node where the failure occurred to the Map Request originator, then
the originator would retry sending the Map Request several times before
timing-out. If such a timeout occurred, then the destination would be
considered not LISP-capable and LISP encapsulation would not be done
toward it.

Dino wrote:

Since Map-Requets are small, it is unlikely that an MTU violation will occur 
for a Map-Request forwarded on the ALT. But if the case did arise, there would 
be fragmentation and reassembly at each GRE end-point since a Map-Request, when 
sent from one ALT node to another, is encapsulated in a GRE tunnel header.

I'm looking at the Map-Request message format and cannot quite convince myself 
that there would never be a situation where MTU would be exceeded. 99.999% of 
the time it would be fine, but I think we need to describe (if not fix) what 
happens when its not. How about adding the following item to the Section 1 
experimental list:

"o  effects of limited possibilities for returning an ICMP message from ALT"

and this text somewhere:

"Given that packets delivered through ALT have an RLOC source address and an EID 
destination address, returning an ICMP error may not always be possible. Since 
Map-Request messages are typically small, it is unlikely that an MTU violation will occur 
for a Map-Request forwarded on the ALT and if it did, it may be possible to deal with it 
through fragmentation of GRE packets. But in the unlikely event of having to generate an 
ICMP error due to MTU or other problems, an ALT node MAY attempt to deliver the error to 
the source RLOC outside ALT. If it turns out to be not possible to deliver the ICMP 
error, the sender of the Map-Request message will retry, and failing again, will time 
out.This could lead to considering the dstination not LISP-capable. Note that ICMP errors 
are more likely with Data Probes, and may lead to the unreported loss of the encapsulated 
data packet."

Or something along those lines.

Jari

_______________________________________________
lisp mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lisp

Reply via email to