> Richard L. Hamilton writes:
> > minimal part being disabling writes to /dev/*mem),
> my main interest is if
> > that would make it possible to install more kernel
> and driver patches safely in multi-user mode,
> > without having to worry about an inconsistent set
> of modules getting loaded prior to reboot.
> 
> Given that interest, I don't think you want to use
> something like
> modflags, because that'd make deferring the reboot
> too painful.
> Instead, we would need a run-time blacklist of
> modules that shouldn't
> be loaded.  The prepatch script would then add an
> entry to the
> blacklist so that the driver can't be loaded (behaves
> as ENOENT) until
> the system is restarted.

More complicated, but less restrictive; sufficient perhaps for
expanded safe online patching _if_ the prepatch files consistently
get it right.  (Just locking down to the currently loaded modules would
be far more restrictive but proof against human error.)
 
> That seems like a fairly reasonable RFE to me, though
> ...
> 
> > Of course in the long run, something like a
> combination of LiveUpgrade with a cloneable,
> > promotable ZFS root would let any maintenance from
> minor patches through upgrades be
> 
> ... yes, a ZFS root likely invalidates the whole
> scheme.
> 
> > deferred until relatively convenient.  But let's
> face it, a lot of folks are going to drag
> > their feet on reloading (esp. if they don't have
> space on their typically dinky pair of internal
> > drives for a spare partition for LiveUpgrade) to
> such a configuration.  I would think this could
> 
> I don't understand that part.  Regardless of what you
> do at this
> point, the user must suffer a reboot in order to
> switch out the kernel
> bits.  Either that reboot causes the run-time
> blacklist to get
> discarded (option A) or it causes us to pick up the
> modified ZFS root
> (option B).
> 
> I think both answers are equivalent in terms of
> feet-dragging.

A lot of systems have two internal drives, on which the OS lives;
and perhaps more stuff if there's room.  Changing the module loading
functionality is a patch or at worst an upgrade; but switching to a zfs
root in a lot of cases would probably only get done as a full-blown
backup of non-OS data (and of config files) and reload, which takes
a lot more down time.  Plenty of systems simply weren't initially set up
with space for an alternate boot environment, esp. with older small disks.
So I think it would take a lot longer to get widespread use of a really
modern approach to safe online upgrades than it would to do some
relatively minor things to make the existing mechanisms safer online in
more cases.  (this from seeing a lot of systems that are still chugging along
running Solaris 8 and with a pair of dinky little mirrored boot drives)
 
 
This message posted from opensolaris.org
_______________________________________________
opensolaris-code mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code

Reply via email to