[lisp] Constructing Map-Replies with overlapping prefixes

[Eric Rescorla <[e-mail]>]

I'm trying to piece together the algorithm in 6833-bis S 5.5
for Map-Replies. The text says:

   A Map-Reply returns an EID-Prefix with a mask-length that is less
   than or equal to the EID being requested.

So, consider the case where an MS has two mappings:

   10/8 -> ETR1
   11/8 -> ETR2

If the Map-Request is for 10.0.0.1, then I would return

   10/8 -> ETR1

Now consider what happens when I have three mappings:

   10/16   -> ETR1
   10.0.2/24 -> ETR3 // New
   11/16   -> ETR2

Now we turn to the remainder of this section:

   When an EID moves out of a LISP site [I-D.ietf-lisp-eid-mobility],
   the database mapping system may have overlapping EID-prefixes.  Or
   when a LISP site is configured with multiple sets of ETRs that
   support different EID-prefix mask-lengths, the database mapping
   system may have overlapping EID-prefixes.  When overlapping EID-
   prefixes exist, a Map-Request with an EID that best matches any EID-
   Prefix MUST be returned in a single Map-Reply message.  For instance,
   if an ETR had database mapping entries for EID-Prefixes:

I'm having trouble parsing this, but what I get out of the example
is that the Map-Response is supposed to contain the EID-prefix that
best matches the request, plus whatever exceptions would be required
to create a correct routing table. So, that would mean in the case
where the Map-Request contains 10.0.0.1, the MS would have to reply
with:

   10/16   -> ETR1
   10.0.2/24 -> ETR3 // New

The first of these is necessary to provide correct routing information
for the requested prefix and the second to avoid providing incorrect
routing information for 10.0.2.1.

Do I have this correct? It seems like the alternative is that the MS
or ETR synthesize a new, more specific prefix. Is that what's intended
instead? The example in S 5.5 suggests otehrwise, but perhaps I am
misunderstanding.

Thanks,
-Ekr