On Sat, May 30, 2009 at 1:31 AM, Alan Altmark <[email protected]> wrote:

> I think it would be safe to say that IBM recommends AGAINST issuing a
> STORE HOST unless you are doing so at the direction of Support Centre.  As
> some have discovered, STORE HOST introduces an "unstable element" (the
> systems programmer? };-) ) into the system, creating risk of an abend.

I found Kris illustrated the problem with his approach himself the
best in the follow-up note where he explained he found even a newer
version of the program that probably was the one adjusted for later VM
releases. Even when you have a program to issue the STORE HOST for you
(and without you knowing it) that program may be wrong or not suitable
in your case. Sure, we all zapped hung users out of the way. But this
is last resort actions that you leave to those who can afford to risk
their job for the good case. But I would expect that the 64-bit
changes would have retired most of these old habits.

One of the vendor applications we ran would change a bit in its VMDBK
(when enabled by a configuration file option). It had class C for
another reason already. When we upgraded to a next z/VM release, the
program did not recognize VM/ESA and assumed we ran VM/SP. Bad things
happened. I've used the dump for some time as training material for
novice dump readers.

The main reason we have SET PRIVCLAS is that we notice the need for
this function, and don't want people to zap in memory for this.
Because when auditing shows you someone modifies real memory, he could
have done anything (well, I recall that when looking at this we did
spot the offset in the address as likely reason for the store host).
For the same reason we have many query commands and don't give anyone
just class E and let them peek around in the machine (let alone that
things get complicated when control blocks are paged).

But then, I don't agree with Kris' view of the world that "MAINT must
be allowed to do anything"  In my view, *if* there is a single userid
that is allowed to do anything under any circumstances (like no
auditing) then use of such a userid should be very much limited to
exceptional cases where normal tools don't work.

Rob

Reply via email to