Am 12.02.2018 um 11:02 hat Daniel P. Berrangé geschrieben:
> On Mon, Feb 12, 2018 at 10:58:31AM +0100, Paolo Bonzini wrote:
> > On 10/02/2018 00:07, John Snow wrote:
> > >> + /* Early check to avoid creating target */
> > >> + if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
> > >> + return;
> > >> + }
> > >> +
> > >> aio_context = bdrv_get_aio_context(bs);
> > >> aio_context_acquire(aio_context);
> > >>
> > >>
> > > What's the implication of the temporarily-extant target node that it
> > > needs to be avoided so strictly?
> > >
> >
> > Creating a file on disk, that no one will ever remvoe. :)
>
> Fortunately libvirt's SELinux policy will probably prevent QEMU creating
> it in the first place :-)
Well, calling drive-mirror without allowing QEMU to create the target
image would be a bit pointless, so I think we can assume that libvirt
did set up the file permission so that QEMU can create it. (Unless
mode=existing is used, but I understand that libvirt doesn't want to
create images with qemu-img, so that doesn't seem to be the case...)
I don't know if libvirt takes care to remove a potentially already
created file if the command then fails, but hopefully it does and the
patch is not actually needed with libvirt.
Kevin
