Am 13.08.2019 um 16:43 hat Max Reitz geschrieben:
> On 13.08.19 13:04, Kevin Wolf wrote:
> > Am 12.08.2019 um 20:11 hat Vladimir Sementsov-Ogievskiy geschrieben:
> >> BDRV_BLOCK_RAW makes generic bdrv_co_block_status to fallthrough to
> >> returned file. But is it correct behavior at all? If returned file
> >> itself has a backing file, we may report as totally unallocated and
> >> area which actually has data in bottom backing file.
> >>
> >> So, mirroring of qcow2 under raw-format is broken. Which is illustrated
> >> by following commit with a test. Let's make raw-format behave more
> >> correctly returning BDRV_BLOCK_DATA.
> >>
> >> Suggested-by: Max Reitz <mreitz@redhat.com>
> >> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> > 
> > After some reading, I think I came to the conclusion that RAW is the
> > correct thing to do. There is indeed a problem, but this patch is trying
> > to fix it in the wrong place.
> > 
> > In the case where the backing file contains some data, and we have a
> > 'raw' node above the qcow2 overlay node, the content of the respective
> > block is not defined by the queried backing file layer, so it is
> > completely correct that bdrv_is_allocated() returns false,like it would
> > if you queried the qcow2 layer directly.
> 
> I disagree.  The queried backing file layer is the raw node.  As I said,
> in my opinion raw nodes are not filter nodes, neither in behavior (they
> have an offset option), nor in how they are generally used (as a format).
> 
> The raw format does not support backing files.  Therefore, everything on
> a raw node is allocated.
> 
> (That is, like, my opinion.)
> 
> >                                          If it returned true, we would
> > copy everything, which isn't right either (the test cases should may add
> > the qemu-img map output of the target so this becomes visible).
> 
> It is right.

So we don't even agree what mirroring the raw node should even mean.

I can the see your point when you say that the raw node has no backing
file, so everything should be copied. But I can also see the point that
the raw node can really just be used as a filter that limits the data
exposed from the qcow2 layer, and you want to keep the copy a COW
overlay over the same backing file.

Both are valid use cases in principle and there is no single right or
wrong.

We don't currently support the latter use case because we have only
sync=full or sync=top, but if you could specify a base node instead, we
could probably suport the case without all of the special-casing filter
nodes and backing file childs.

You would call bdrv_co_block_status_above() with the right base node and
it would just recurse whereever the data is stored, be it bs->backing,
bs->file or even driver-specific children. This would allow you to find
out whether some block in the top node came from the base node that
we're going to keep. If yes, skip it; if no, copy it.

This way we wouldn't have to decide whether raw is a filter or not,
because it wouldn't make a difference. The behaviour would only depend
on the base node given by the user. If you specified the top-level qcow2
file as the base, you get your full copy; if you specified the backing
qcow2, you get the partial copy where the target still uses the same
backing file.

(Hm... It would only actually work if the offsets stay the same in the
chain, which is true for backing file children, but not necessarily for
other children. Anyway, even if we don't gain much functionality, I
really want a more generic model that avoids different types of nodes
and edges as much as possible.)

Kevin