On 07/07/2011 12:30 PM, Eric Blake wrote:
> Yes, the ESX
> driver has no means to support a truly read-only connection as you
> always need to provide credentials, so one can consider this a bugfix
> or a regression depending on the point-of-view.
I'm 50-50 on whether this really is a bug-fix, whether it means we
should allow (and otherwise ignore) the flag, as in:
virCheckFlags(VIR_CONNECT_RO, VIR_DRV_OPEN_ERROR);
Anyone else with an opinion?
At this point, I'm making the change to allow (but otherwise ignore)
VIR_CONNECT_RO for the flags cleanup patch. That way, there are no
semantic changes. Beyond that, if we decide that a particular driver
needs to pay attention to a flag, then we can later make a followup
patch that restricts the flag for just that driver.
Remember, read-only connections have already filtered out API calls from
ever reaching driver callbacks in the first place. So, while it is
conceivable that some drivers may need to do something different to
connect to their hypervisor on a read-only connection, it appears that
most (all?) drivers are just fine ignoring that piece of information
from the connection (they behave the same whether read-only or
read-write, for the callbacks that can actually be triggered on a
read-only connection).
--
Eric Blake eblake(a)redhat.com +1-801-349-2682
Libvirt virtualization library
http://libvirt.org