Re: [DNSOP] WGLC for draft-ietf-dnsop-let-localhost-be-localhost-02

[Viktor Dukhovni <[e-mail]>]

On Fri, Jan 26, 2018 at 08:22:18AM -0800, 神明達哉 wrote:

> Hmm, that's different from my interpretation of the draft.  According
> to my usual interpretation of IETF docs, I would interpret these from
> Section 3:
> 
>    3.  Name resolution APIs and libraries MUST recognize localhost names
>        as special, and MUST always return an appropriate IP loopback
>        address for IPv4 and IPv6 address queries and negative responses
>        for all other query types.  Name resolution APIs MUST NOT send
>        queries for localhost names to their configured recursive DNS
>        server(s).
> 
>        As for application software, name resolution APIs and libraries
>        MUST NOT use a searchlist to resolve a localhost name.
> 
>    4.  (Caching) recursive DNS servers MUST respond to queries for
>        localhost names with NXDOMAIN.
> 
>    5.  Authoritative DNS servers MUST respond to queries for localhost
>        names with NXDOMAIN.
> 
> as these are requirements without a user-configurable knob.  If the
> actual intent was just to specify the default behavior with a
> configurable knob, I'd expect SHOULD-variants are used in cases like
> these.

Exactly.  The MUST language is asking implementations to remove
support for the existing knobs, which should stay.

I just talked to Christos Zoulas of NetBSD, as he's also the current
(or at least recent) upstream maintainer of libresolv.  He not only
supports my view that the MUST is extraneous, but indeed also has
localhost (forward) and loopback (reverse) zones configured on his
own machines.  This practice is reasonably widespread.

I should also note that there is some confusion in recent messages
between what is a stub (DNS) resolver and higher-level name resolution
facilities such as getaddrinfo(3).  Yes, getaddrinfo(3) and friends
on modern Unix-like systems use nsswitch and typically consult the
/etc/hosts file before using DNS, but that's not what this draft
is about.  It specifically describes "stub-resolver" behaviour, not
getaddrinfo() behaviour.  The stub resolver is libresolv, and does
not presently special-case "localhost".

If the intent is to require special handling of "localhost" in the
platform's name to address lookup library (getaddrinfo(),
gethostbyname(), ...), then the draft should say so, instead of
talking about stub resolvers, which are only the DNS component of
the platform's hostname resolution stack.

-- 
	Viktor.