On Mon, Aug 1, 2011 at 11:26 AM, Kevin Wolf <kw...@redhat.com> wrote: > Am 29.07.2011 06:49, schrieb Devin Nakamura: >> Still in very raw form. Not likely to work yet >> >> Signed-off-by: Devin Nakamura <devin...@gmail.com> > > I don't think it's quite as easy. > > The problem is that qcow2 will access the image header in some places, > for example when growing the L1 or refcount table. You need to make sure > that it doesn't destroy the source format header, but writes to the > target format header at a different offset. This is why I've made sure to size the L1 and refcount tables so they wont need to be resized. But I suppose that's really a balancing act that's likely to break if someone makes changes to the current code, or when snapshots are implemented
> I think much of qcow2_open and qcow2_open_conversion_target could be the > same. That is, both could call a common function that takes a parameter > for the header offset. I'm almost certain I could do that (considering I lifted a lot of the code for qcow2_open_conversion_target from qcow2_open) > The other part of qcow2_open_conversion_target (creating a new header > and a basic L1/refcount table) could possibly be shared with > qcow2_create2, though I'm not quite as sure about this one. >> + //TODO: double check this, it seems a bit fishy > > That's my impression of the whole operation, too. :-) > > This is why I thought it would be nice to share this code with normal > image creation. Then we have this fishy code at least only once. > > Kevin >
