All about CNAME record

Kevin Darcy kcd at daimlerchrysler.com
Wed May 31 17:50:04 UTC 2000


mgourite at my-deja.com wrote:

>   Hello,
>
>   I would like to have informations about the CNAME record :
>
>   1) Why we can't associate an alias at NS, SOA, CNAME, MX records ?
> It's because, it's recommended by the RFC or it's a piece of advice to
> simplify DNS administrator's life.

Today, it is generally considered *mandatory* for labels in the RDATA of
an RR to be canonical names rather than aliases, even though RFC 1034
uses the term "should" -- back when 1034 was written, the IETF wasn't
quite so picky about requirements-levels terms, i.e. should/must.
RFC 2119 nailed all of this down eventually.

The main rationale for the _de_facto_ mandate? Efficiency. CNAMEs
increase packet size and require more nameserver -- and in some cases
application -- processing than canonical names. While they may have valid
uses at an application level, i.e. it may be desirable for users to know
the same server by multiple names, there really is no legitimate reason
to use them in zonefile RDATA's, which are mostly transparent to the
users and often maintained programmatically. So, in the absence of any
visible benefit, do things the efficient way.

>   2) What are the problems generated by this use of CNAME ? I heard
> about problems with MTA (sendmail), old BIND versions, ...

Yes, when following the RFC 974 logic for MX record processing, mailers
need to identify "self" references in a given MX list in order to prevent
loops. This is greatly complicated if CNAMEs are allowed as MX targets.
Note that sendmail is not the only mailer that is impacted by this:
*any* mailer that uses MX records is going to have a tougher time
discerning "self" references if CNAMEs are permitted as MX targets. The
problem is not implementation-specific.

You are certainly welcome to lobby for a relaxation of this _de_facto_
mandate, but I think the "namedroppers" list is probably a better forum
for that than "bind-users", since it's not at all BIND-specific.


- Kevin





More information about the bind-users mailing list