Re: [Qemu-devel] [PATCH v2 2/2] block: move the bdrv_dev_change_media_cb()

[Luiz Capitulino]

On Mon, 17 Jun 2013 16:49:11 +0200
Kevin Wolf <address@hidden> wrote:

> Am 17.06.2013 um 15:51 hat Luiz Capitulino geschrieben:
> > On Mon, 17 Jun 2013 15:46:52 +0200
> > Kevin Wolf <address@hidden> wrote:
> > 
> > > Am 17.06.2013 um 15:38 hat Pavel Hrdina geschrieben:
> > > > >>>>>It's just a warning, that you used a password for a block device 
> > > > >>>>>that
> > > > >>>>>doesn't require it. The device is opened successfully and should be
> > > > >>>>>handled correctly (call the bdrv_dev_change_media_cb() ).
> > > > >>>>
> > > > >>>>Yep, IMO it's worth a comment that this isn't an "error" just a
> > > > >>>>"warning".
> > > > >>>
> > > > >>>Actually, you can't have such a warning in QMP. You either fail or 
> > > > >>>you
> > > > >>>succeed. We should just do what the current code does.
> > > > >>>
> > > > >>
> > > > >>This is the same logic as the old one. The device is loaded but the
> > > > >>error is emitted.
> > > > >
> > > > >That's a bug if the operation succeeded.
> > > > >
> > > > 
> > > > In that case, how do you think, that we should handle the situation
> > > > that user is trying to open device that isn't require the password, but
> > > > user will provide the password?
> > > > 
> > > > I don't think that we should fail and abort that operation.
> > > 
> > > I think we should. The image and the options passed for it don't fit
> > > together, this is an error condition. Probably the user meant to pass a
> > > different image.
> > 
> > I agree in principle, but I fear this might be an incompatible change as
> > there might be clients out there assuming the VM is up and running (because
> > it's what ends up happening).
> > 
> > Thinking about this again though, the client does get an error...
> 
> Do you think any client is sending passwords for unencrypted images?
> Because if there is none (and I think we have reason to believe so), we
> don't break anything if we change the behaviour. And if something
> does break, we have uncovered a management tool bug, so that's not too
> bad either.

Yes, I agree. I was being overly cautious when I suggested dropping the
error, but I think you're right: we do send an error, so a well written
client should just fail and shouldn't brake if we do the right thing.

So let's do the Right Thing, but I also suggest to do this in a separate
commit so that it's easy to spot.