Question regarding recursion...

[Cricket Liu]

> In an attempt to avoid scanning the RFC's, I'm hoping someone knows
off-hand
> whether the following behavior exhibited by a vendor's (who shall remain
> nameless) implementation of BIND 8.2.2P7 is proper (since it appears to
break
> some stub resolvers) :
>
> A stub resolver issues a recursive query request to said above - the said
> above (even with recursion supposedly enabled) returns a query response
(with
> the appropriate RR info, not a referral) with the "recursion NOT
available"
> bit set for any RR's the server is either authoritative for itself or that
it
> has cached - otherwise, it returns the query response with the "recursion
> available" bit set for any RR's that it actually would need to start the
query
> process for. I guess in a sense this has some logic, since there was no
> further querying needed in the former case....
>
> Another vendor's (also remain nameless) implementation of BIND 8.2.3 does
not
> exhibit the above behavior - i.e. returns all query responses with the
> "recursion available" bit set.
>
> Which vendor is following proper basic BIND recursive procedure (if there
is
> any) ???

That's bizarre.  Here's what RFC 1035 has to say about the RA bit:

RA              Recursion Available - this be [sic] is set or cleared in a
                response, and denotes whether recursive query support is
                available in the name server.

So RA indicates whether recursive query support is *available*, not
whether recursion was used.  I'd argue that clearing it for responses
that didn't require recursion is a misinterpretation of the RFC.

cricket