On Tue, Jan 14, 2020 at 04:43:36PM +0100, Max Reitz wrote:
> On 09.01.20 12:10, Stefan Hajnoczi wrote:
> > Add qemu-img measure support in the "luks" block driver.
> > 
> > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> > ---
> >  block/crypto.c | 82 ++++++++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 82 insertions(+)
> > 
> > diff --git a/block/crypto.c b/block/crypto.c
> > index ed32202fa2..51f37bb1f6 100644
> > --- a/block/crypto.c
> > +++ b/block/crypto.c
> > @@ -548,6 +548,87 @@ static int64_t block_crypto_getlength(BlockDriverState *bs)
> 
> [...]
> 
> > +            if (ret & BDRV_BLOCK_ZERO) {
> > +                /* Skip zero regions */
> > +            } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
> > +                       (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
> > +                /* Count clusters we've seen */
> > +                required += pnum;
> > +            }
> 
> Don’t LUKS-encrypted files allocate effectively everything because zero
> data has to be encrypted, too?
> 
> (“Effectively”, because you could zero out regions that are zero when
> encrypted, but...)
> 
> > +        }
> > +    }
> > +
> > +    /* Take into account preallocation.  Nothing special is needed for
> > +     * PREALLOC_MODE_METADATA since metadata is always counted.
> > +     */
> > +    if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
> > +        required = virtual_size;
> 
> Same here.  I think required should always be set to virtual_size.

Good points.  I may have inherited this from the qcow2 code, where the
L2 tables can still have unallocated/zero clusters.

I'll check if this logic makes sense outside of qcow2.

Stefan