Re: [Qemu-devel] [PATCH 02/10] RFC: Implement qcow2's snapshot dependent saving function.

[Stefan Hajnoczi]

On Tue, May 08, 2018 at 09:50:00AM -0500, Eric Blake wrote:
> On 03/13/2018 03:33 AM, address@hidden wrote:
> > From: Junyan He <address@hidden>
> > 
> > For qcow2 format, we can increase the cluster's reference count of
> > dependent snapshot content and link the offset to the L2 table of
> > the new snapshot point. This way can avoid obvious snapshot's dependent
> > relationship, so when we delete some snapshot point, just decrease the
> > cluster count and no need to check further.
> > 
> > Signed-off-by: Junyan He <address@hidden>
> > ---
> >   block/qcow2-snapshot.c | 154 
> > +++++++++++++++++++++++++++++++++++++++++++++++++
> >   block/qcow2.c          |   2 +
> >   block/qcow2.h          |   7 +++
> >   3 files changed, 163 insertions(+)
> 
> It sounds like you are trying to modify the qcow2 spec to store more
> information into the internal snapshot table (and if you aren't, why not? If
> an internal snapshot depends on another one and we don't record that
> information in the qcow2 metadata, then we are pushing the burden of
> tracking inter-relationships onto management apps, and risk corruption if
> you load a snapshot without also visiting its dependencies).  It is
> absolutely essential that such modifications be reflected in
> docs/interop/qcow2.txt first, to make sure we agree on the implementation.
> What's more, there is already another parallel proposal that is also wanting
> to tweak qcow2 files:
> 
> https://lists.gnu.org/archive/html/qemu-devel/2018-04/msg05231.html
> 
> We need to make sure both additions are coordinated.

I think the relationship doesn't need to be stored.  The point of this
snapshot clone operation is to share some of the clusters of the parent
snapshot (using qcow2's existing cluster reference counts).  Cloning
avoids having to write a duplicate copy of those clusters.

Stefan

signature.asc
Description: PGP signature