Re: [Libguestfs] [libguestfs PATCH] guestfs.pod: document encrypted RBD disk limitation
On 05/18/22 11:30, Richard W.M. Jones wrote:
On Wed, May 18, 2022 at 10:30:14AM +0200, Laszlo Ersek wrote:
> Under "REMOTE STORAGE", the "NETWORK BLOCK DEVICE" section
already
> documents some limitations. Turns out we need to describe a quirky
> exception for accessing encrypted RBD disks, too.
>
> Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2033247
> Signed-off-by: Laszlo Ersek <lersek(a)redhat.com>
> ---
> lib/guestfs.pod | 33 ++++++++++++++++++++
> 1 file changed, 33 insertions(+)
>
> diff --git a/lib/guestfs.pod b/lib/guestfs.pod
> index b04c28d62750..1ad44e7c2878 100644
> --- a/lib/guestfs.pod
> +++ b/lib/guestfs.pod
> @@ -679,6 +679,39 @@ servers. The server string is documented in
> L</guestfs_add_drive_opts>. The C<username> and C<secret>
parameters are
> also optional, and if not given, then no authentication will be used.
>
> +An encrypted RBD disk -- I<directly> opening which would require the
> +C<username> and C<secret> parameters -- cannot be accessed if the
> +following conditions all hold:
> +
> +=over 4
> +
> +=item *
> +
> +the L<backend|/BACKEND> is libvirt,
> +
> +=item *
> +
> +the image specified by the C<filename> parameter is different from the
> +encrypted RBD disk,
> +
> +=item *
> +
> +the image specified by the C<filename> parameter has L<qcow2
> +format|/COMMON VIRTUAL DISK IMAGE FORMATS>,
> +
> +=item *
> +
> +the encrypted RBD disk is specified as a backing file at some level in
> +the qcow2 backing chain.
> +
> +=back
> +
> +This limitation is due to libvirt's (justified) separate handling of
> +disks vs. secrets. When the RBD username and secret are provided inside
> +a qcow2 backing file specification, libvirt does not construct an
> +ephemeral secret object from those, for Ceph authentication. Refer to
> +L<https://bugzilla.redhat.com/2033247>.
> +
> =head3 FTP, HTTP AND TFTP
Seems fine. I had to look at perlpod(1) to check that the piped L<>
references were correct, but they seem to be!
Yeah, I recalled both piped-but-https-references, and non-piped-local
references, from the PODs, so I figured I'd try piped-local -- and the
rendered HTML (which I reviewed in Firefox) confirmed they were going to
work!
Acked-by: Richard W.M. Jones <rjones(a)redhat.com>
Rich.
Thanks!
Laszlo