On Wed 14 May 2008 at 08:42AM, james hughes wrote:
> 
> On May 14, 2008, at 2:13 AM, Dan Price wrote:
> 
> >On Tue 13 May 2008 at 03:00PM, Gary Winiger wrote:
> >>This project enables a policy where "root" can never be used  
> >>directly by
> >>administrators as an account providing full system access.
> >>In a Major release this policy may be made the default.
> >
> >To be clear: I'm agnostic about the goodness of this idea; I'm  
> >concerned that the defintion of "default" is not clear, and that the
> >zones case has not been fully explored here.
> 
> I hope the previous message helps with this...

Well it does in the sense that it provides the motivation and
context.  I was really trying to point out something different, but wasn't
clear.  It also doesn't answer my questions about sys-unconfig.
To back up slightly to my confusion: The text of the case says:

> In a Major release this policy may be made the default.

This statement is vague-- if this case is approved, does that mean
that a team working on a major release may make this the default
without further review?

Or does it mean that in the event of a major release, PSARC would
see another case which effects this change?

If the latter, then please rewrite:

"This case does not set this policy as the default for the system.  A
future major release could adopt this policy as the default, although
further architectural change will be required in order to support
non-global zones and to alter existing installation technologies
(if supported) to properly provision non-root users.  Future cases
will address this."

There is a new requirement being placed on the installation
technologies here-- when this policy is active, then delivering
/etc/passwd from packages is no longer enough to provision the system--
and so when this policy is the default, wider changes are going to
be needed.  And that is what I think needs resolution here.  There should
be a clear specification of what an installer should do-- otherwise,
you'll see all sorts of bizarre stuff happening with packages and
install-time provisioning (as we saw in DP1 and DP2).

> >In a sense, a zone is the "most pure"
> >version of this problem, since it doesn't get much provisioned by an
> >installer, really--  it just gets laid out on the system, and lets
> >sysidtool do the rest.
> 
> Agreed that the zone issue has not been fully explored and we should  
> do this before approving this case. I did create a zone on DP1 and ran  
> into this bug, and I seem to remember using "zlogin -S" or something  
> like that to get around the lack of a root password. Fixing this is  
> mandatory.

Great!

As an aside, DP1 zones support (and 2008.05 zones support for that matter)
isn't really representative of anything other than a hack and slash
attempt to get *something* working in the allotted time.  It should not be
regarded as architecture.

For further background on how this issue has impacted zones,
you may wish to review:

http://defect.opensolaris.org/bz/show_bug.cgi?id=681

        -dp

-- 
Daniel Price - Solaris Kernel Engineering - dp at eng.sun.com - blogs.sun.com/dp

Reply via email to