On Tue, Aug 18, 2020 at 11:47 AM Kevin Wolf <kwolf@redhat.com> wrote:

> Am 17.08.2020 um 19:19 hat Nir Soffer geschrieben:
> > On Thu, Aug 13, 2020 at 7:36 PM Kevin Wolf <kwolf@redhat.com> wrote:
> >
> > > Instead of implementing qemu-nbd --offset in the NBD code, just put a
> > > raw block node with the requested offset on top of the user image and
> > > rely on that doing the job.
> > >
> > > This does not only simplify the nbd_export_new() interface and bring it
> > > closer to the set of options that the nbd-server-add QMP command
> offers,
> > > but in fact it also eliminates a potential source for bugs in the NBD
> > > code which previously had to add the offset manually in all relevant
> > > places.
> > >
> >
> > Just to make sure I understand this correctly -
> >
> > qemu-nbd can work with:
> >
> >     $ qemu-nbd 'json:{"driver": "file", "filename": "test.raw"}'
> >
> > And:
> >
> >     $ qemu-nbd 'json:{"driver": "raw", "file": {"driver": "file",
> > "filename": "test.raw"}}'
> >
> > I assumed that we always create the raw node?
>
> No, the first form creates only the 'file' node without a 'raw' node on
> top. For all practical matters, this should be the same in qemu-img or
> qemu-nbd. For actually running VMs, omitting the 'raw' node where it's
> not needed can improve performance a little.
>

We did not check if we have different performance with the extra raw node.
Since in our use case (copying images) small reads/writes are unlikely, I
don't
think it will make a difference.

What is true is that if you use a filename without specifying the driver
> (i.e.  you rely on format probing), you'll get a 'raw' node on top of
> the 'file' node.
>
> > oVirt always uses the second form to make it easier to support offset,
> > size, and backing.
> >
> https://github.com/oVirt/ovirt-imageio/blob/2021164d064227d7c5e03c8da087adc66e3a577e/daemon/ovirt_imageio/_internal/qemu_nbd.py#L104
> >
> > This also seems to be the way libvirt builds the nodes using -blockdev.
>
> libvirt actually has a BZ to avoid the 'raw' node for performance when
> it's not needed.
>
> > Do we have a way to visualize the internal node graph used by
> > qemu-nbd/qemu-img?
>
> No, but as long as you explicitly specify the driver, you get exactly
> what you specified.
>

So this is not really needed then.


> For exploring what happens, you can pass the same json: filename to QEMU
> (maybe with -hda) and then use the monitor to inspect the state.
>
> Kevin
>
>