Re: [Qemu-devel] Re: KVM call agenda for July 27

[Daniel P. Berrange]

On Tue, Jul 27, 2010 at 10:28:04AM -0500, Anthony Liguori wrote:
> On 07/27/2010 10:22 AM, Markus Armbruster wrote:
> >Kevin Wolf<address@hidden>  writes:
> >
> >   
> >>Am 27.07.2010 15:00, schrieb Anthony Liguori:
> >>     
> >>>On 07/27/2010 02:19 AM, Markus Armbruster wrote:
> >>>       
> >>>>Anthony Liguori<address@hidden>   writes:
> >>>>
> >>>>
> >>>>         
> >>>>>- any additional input on probed_raw?
> >>>>>
> >>>>>           
> >>>>Isn't it a fait accompli?  I stopped providing input when commit
> >>>>79368c81 appeared.
> >>>>
> >>>>         
> >>>No.  79368c81 was to close the security hole (and I do consider it a
> >>>security hole).  But as I mentioned on the list, I'm also not satisfied
> >>>with it and that's why I proposed probed_raw.  I was hoping to get a
> >>>little more input from those that objected to 79368c81 as to whether
> >>>probed_raw was more agreeable.
> >>>       
> >>Actually I believe qraw is less agreeable. It just too much magic. You
> >>wouldn't expect that your raw images are turned into some other format
> >>that you can't mount or use with any other program any more.
> >>     
> >I also dislike probed_raw, for the same reasons.
> >
> >Raw can't be probed safely, by its very nature.  For historical reasons,
> >we try anyway.  I think we should stop doing that, even though that
> >breaks existing use relying on the misfeature.  Announce it now, spit
> >out scary warnings, kill it for good 1-2 releases later.
> >
> >If we're unwilling to do that, then I'd *strongly* prefer doing nothing
> >over silently messing with the raw writes to sector 0 (so does
> >Christoph, and he explained why).
> 
> If we add docs/deprecated-features.txt, schedule removal for at least 1 
> year in the future, and put a warning in the code that prints whenever 
> raw is probed, I think I could warm up to this.
> 
> Since libvirt should be insulating users from this today, I think the 
> fall out might not be terrible.

Next libvirt (0.8.3)  will always set format to 'raw' if the user does
not give any alternative. Thus QEMU's probing code will never be run
for the main disk. We can't stop  it probing for backing stores, with
the exception of qcow2 images using the embedded backingstore format.

BTW, it is questionable whether VMDK should be probing for backing
store format all, rather than forcing all VMDK backing stores to
also be VMDK. (If we want  full compliance/compatbility with VMWare's
impl which obviously doesn't use stuff like qcow2)

For cow & qcow perhaps we should mark them as deprecated and explicitly
recommend people to use qcow2 instead ? Or make them only use backing
stores whose format matches the main store ?

Daniel
-- 
|: Red Hat, Engineering, London    -o-   http://people.redhat.com/berrange/ :|
|: http://libvirt.org -o- http://virt-manager.org -o- http://deltacloud.org :|
|: http://autobuild.org        -o-         http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505  -o-   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|