SCIENTIFIC-LINUX-USERS Archives

May 2012

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Steve Rikli <[log in to unmask]>
Reply To:
Steve Rikli <[log in to unmask]>
Date:
Thu, 3 May 2012 17:47:57 -0700
Content-Type:
text/plain
Parts/Attachments:
text/plain (82 lines)
On Thu, May 03, 2012 at 05:04:09PM -0700, Konstantin Olchanski wrote:
> On Thu, May 03, 2012 at 03:10:01PM -0700, Steve Rikli wrote:
> > 
> > By chance do your NFS server & NFS clients have different DNS domainnames
> > configured?
> 
> I am not sure there is such a thing as a "DNS domainname".

I agree there are shades of grey depending on which OS one is talking
about.

So for clarity, in this instance and in my earlier example, when I say
"DNS domainname" I am referring to the output of 'dnsdomainname' on my
Linux desktop; which is also the same as my FQDN minus the leading
hostname up to the first "." .

E.g. "foo.bar.domain.com" is 'hostname', and thus "bar.domain.com" is
'dnsdomainname'.  Hopefully that is clear.

> There is many confusions - i.e. some people think that the command
> "domainname" returns the domain name - when in fact it returns the NIS domain
> name.

... and in some environments, the NIS and DNS domains use the same value,
furthering the misconception.  :-)

> So there is: a) the NIS domain name (returned by "domainname", should have
> been named "ypdomainname").

In some Linux distributions, it (almost) is:  'nisdomainname' .

> Where is the
> value of "hostname" come from? Good luck finding out. At TRIUMF it comes from
> DHCP, set randomly either to "foo" and "foo.triumf.ca". Also can be set in
> /etc/sysconfig/network.

Indeed.  It depends on your situation and your environment.

> > Or, maybe the NFS server simply defines a different NFSv4 domainname than
> > your clients' default DNS domainname?
> 
> Right. Default DNS domainname, NFSv4 domainname. I guess these terms are
> clearly defined in some RFC somewhere. I know not what they mean. (See
> above).

Okay.  Hopefully my explanation and example makes the behavior clearer.

FWIW, rpc.idmapd(8) also has this:

  -d domain    Set domain to domain.  This is used internally by NFSv4 and
               is typically assigned by the system administrator.  By
               default, domain is set to be the FQDN of the host, minus the
               hostname.

Which is essentially echoed by the comments in some distributions' version
of /etc/idmapd.conf , e.g.:

   # The following should be set to the local NFSv4 domain name
   # The default is the host's DNS domain name.
   #Domain = local.domain.edu

> > Whether the NFSv4 domain is defined separately, or the default value of
> > DNS domainname is used, both client & server must have the same setting.
> > IME when ownership comes up with nobody.nobody it's usually a mis-match
> > between client & server NFSv4 domain names.  Or rpc.idmapd is dead or
> > otherwise out in the weeds.  :-)
> 
> I confirm this. NFSv4 server and client have to use the same domain name,
> and by luck, idmapd has a debug mode where it complains about mismatched
> usernames and does report what domain names it is trying to use.
> 
> As I said before, SL6.2 machines part of an NIS cluster seem to get
> the NFSv4 domain name correctly every time automatically. Good.

Most likely then, 'dnsdomainname' on those SL6.2 machines is the same
as 'dnsdomainname' on the NFSv4 server.  And in the absence of a
definition for "Domain=" in /etc/idmapd.conf , that is what is being
used for the NFSv4 domain.

Cheers,
sr.

ATOM RSS1 RSS2