On 26.10.2015 19:12, John Snow wrote:
> For more complex BDS trees that can be created under normal circumstances,
> we lose the ability to issue query commands because of our inability to
> re-construct the absolute filename.
> 
> Instead, omit this field when it is a problem and present as much information
> as we can.
> 
> This will change the expected output in iotest 110, where we will now see a
> json filename and the lack of an absolute filename instead of an error.
> 
> Signed-off-by: John Snow <jsnow@redhat.com>
> ---
>  block/qapi.c               | 10 ++++++----
>  tests/qemu-iotests/110.out |  5 ++++-
>  2 files changed, 10 insertions(+), 5 deletions(-)

One problem I see now is that qemu-img --backing-chain uses the full
backing name if it is available and if it isn't, it resorts to the
non-full backing name (which was good until this patch, as far as I can
see). But now that's not necessarily a valid substitution anymore:

$ mkdir foo
$ ./qemu-img create -f qcow2 foo/base.qcow2 64M
$ ./qemu-img create -f qcow2 -b base.qcow2 foo/top.qcow2
$ ./qemu-img create -f qcow2 base.qcow2 64M
$ ./qemu-img info --backing-chain foo/top.qcow2
image: foo/top.qcow2
file format: qcow2
virtual size: 64M (67108864 bytes)
disk size: 196K
cluster_size: 65536
backing file: base.qcow2 (actual path: foo/base.qcow2)
Format specific information:
    compat: 1.1
    lazy refcounts: false
    refcount bits: 16
    corrupt: false

image: foo/base.qcow2
file format: qcow2
virtual size: 64M (67108864 bytes)
disk size: 196K
cluster_size: 65536
Format specific information:
    compat: 1.1
    lazy refcounts: false
    refcount bits: 16
    corrupt: false
$ ./qemu-img info --backing-chain \
"json:{'file.filename':'foo/top.qcow2',
       'driver':'qcow2','lazy-refcounts':true}"
image: json:{"lazy-refcounts": true, "driver": "qcow2", "file":
{"driver": "file", "filename": "foo/top.qcow2"}}
file format: qcow2
virtual size: 64M (67108864 bytes)
disk size: 196K
cluster_size: 65536
backing file: base.qcow2
Format specific information:
    compat: 1.1
    lazy refcounts: false
    refcount bits: 16
    corrupt: false

image: base.qcow2
file format: qcow2
virtual size: 32M (33554432 bytes)
disk size: 196K
cluster_size: 65536
Format specific information:
    compat: 1.1
    lazy refcounts: false
    refcount bits: 16
    corrupt: false

As you can see, in the second case, the wrong base.qcow2 was used. I
think qemu-img info --backing-chain should abort once it hits a point
where has_full_backing_filename is false; but for that to work, we need
to set info->full_backing_filename even if the "relative" and the
absolute backing filename (backing_filename and backing_filename2) are
equal.

(By the way: It's actually some kind of a bug that you can open images
with JSON filenames and backing files; this only works because
bdrv_open_backing_file() is called before bdrv_refresh_filename().
Actually, after my "Drop BDS.filename" series it will no longer work.
This is bad. I need to fix this before I can continue the series...)

((It is bad because that means:
$ x86_64-softmmu/qemu-system-x86_64 \
  -drive if=none,file.filename=foo/top.qcow2,lazy-refcounts=on
qemu-system-x86_64: -drive
if=none,file.filename=foo/top.qcow2,lazy-refcounts=on: Cannot use
relative backing file names for 'json:{"lazy-refcounts": "on", "driver":
"qcow2", "file": {"driver": "file", "filename": "foo/top.qcow2"}}'

And that's bad.))

(((So for this patch that means I guess we should just fix
bdrv_get_full_backing_filename() instead.)))

Max