Empty CNAME chain, should getaddrinfo() return EAI_NONAME or EAI_FAIL?

[Chuck Swiger]

On Apr 28, 2011, at 11:52 AM, Doug Barton wrote:
>> Agreed.  Akamai's EdgeSuite doesn't provide IPv6 AAAA records at this time, but e3191.c.akamaiedge.net does have an A record.
> 
> I understand what you're saying, but I've always referred to such a thing as an "empty CNAME chain" because it doesn't result in an address record at the end.  Is there a more proper term for it?

RFC-1034 talks about CNAME chains: 

"Of course, by the robustness principle, domain software should not fail when presented with CNAME
chains or loops; CNAME chains should be followed and CNAME loops signalled as an error.

...and the recursive query algorithm:

"If recursive service is requested and available, the recursive response
to a query will be one of the following:

   - The answer to the query, possibly preface by one or more CNAME
     RRs that specify aliases encountered on the way to an answer.

   - A name error indicating that the name does not exist.  This
     may include CNAME RRs that indicate that the original query
     name was an alias for a name which does not exist.

   - A temporary error indication."

This phrase "name does not exist" appears to best be represented as EAI_NONAME.

>>>> should getaddrinfo() return EAI_NONAME or EAI_FAIL?
>>> 
>>> RFC 3493 says:
>>> 
>>>   [EAI_NONAME]    The name does not resolve for the supplied
>>>                   parameters.  Neither nodename nor servname were
>>>                   supplied.  At least one of these must be supplied.
>>> 
>>>   [EAI_FAIL]      A non-recoverable error occurred when attempting to
>>>                   resolve the name.
>>> 
>>> Which means that it should probably return EAI_NONAME; it's the least
>>> bad error code among the ones listed in RFC 3493 for getaddrinfo(),
>>> although one should not be mislead to think that this means that the
>>> DNS said NXDOMAIN.
>> 
>> +1 to this analysis as well.
> 
> The original question that started me down the rabbit hole was, "What error code should 'ping6 www.apple.com' return?"

Well, it appears that some other folks expect EAI_NONAME:

  http://www.freebsd.org/cgi/query-pr.cgi?pr=156684

...and the MacOSX version of ping6 uses it for this case:

% ping6 www.apple.com
ping6: getaddrinfo -- nodename nor servname provided, or not known

> What confuses me here is that the node does actually "exist" in the sense that there is _an_ address record for it. So attempting to look at this from the standpoint of a user, the error message I get back (in our case, "hostname or servname not provided, or not known") doesn't make any sense. (Although admittedly the "does not resolve for the supplied parameters" part of the definition does seem to.) Since for the purpose of ping6 no AAAA record at the end of the chain is a non-recoverable error, _FAIL ("non-recoverable failure in name resolution") seems to make more sense.

If the nameserver returned a failure (ie, NXDOMAIN, SERVFAIL), then I would agree with EAI_FAIL.

However, it didn't fail; you just asked a question which it answered successfully by providing all of the data which matched the query type; ie, the CNAMES up to the end of the chain, but not including the A record which terminates it, since you asked for an AAAA instead.

> Is it possible that what we need is a new error code to designate "something exists here, just not what you asked for?" If _NONAME is intended to indicate that it seems overloaded. Alternatively I think we need to improve the language of our error messages. :-/

As Harvard mentioned, EAI_NODATA?  :-)

Regards,
-- 
-Chuck