[Libguestfs] [PATCH v2v] convert: Ignore /dev/mapper/osprober-* devices when trimming

[Laszlo Ersek]

On 05/16/22 17:46, Richard W.M. Jones wrote:
> These devices can be left around by either grub2 or the osprober tool.
> They are read-only mirrors of existing filesystems and it appears we
> can safely ignore them.
> 
> Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=2003503
> ---
>  convert/convert.ml | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/convert/convert.ml b/convert/convert.ml
> index 87fca7252b..997f6b08bd 100644
> --- a/convert/convert.ml
> +++ b/convert/convert.ml
> @@ -194,10 +194,16 @@ and do_fstrim g inspect =
>    (* Get all filesystems. *)
>    let fses = g#list_filesystems () in
>  
> +  (* Ignore unknown/swap devices. *)
>    let fses = List.filter_map (
>      function (_, ("unknown"|"swap")) -> None | (dev, _) -> Some dev
>    ) fses in
>  
> +  (* Ignore filesystems left around by osprober (RHBZ#2003503). *)
> +  let fses =
> +    List.filter (fun dev -> not (String.is_prefix dev "/dev/mapper/osprober-"))
> +                fses in
> +
>    (* Trim the filesystems. *)
>    List.iter (
>      fun dev ->
> 

I've read up on os-prober:

https://www.thegeekdiary.com/how-to-disable-os-prober-in-centos-rhel-7/
https://github.com/campadrenalin/os-prober
http://joeyh.name/code/os-prober/

and it's *absolutely infuriating* that os-prober itself, or whatever
invokes os-prober, litters the system with block device nodes. In fact,
I don't even know how it is *possible* for a device node to be left
around under /dev/mapper -- I thought /dev/mapper/ would be re-populated
*from zero* every time the system boots? Do we have something in the
appliance perhaps that *creates* these files?

I'm quite unhappy about ignoring these filesystems in "do_fstrim"
*only*. If guestfs_list_filesystems() returns them in the first place,
won't that cause confusion for other libguestfs applications or scripts?
Do we expect the same filesystem to appear multiply in the
list-filesystems output?

I think this is a serious bug in some component different from
libguestfs / virt-v2v.

Hmmm... After reading your comment
<https://bugzilla.redhat.com/show_bug.cgi?id=2003503#c12>, I think we
absolutely need to report a bug for grub2. I'll contact the rhboot team.

I don't disagree that we have to work it around, but until we understand
the problem better, I don't feel safe about filtering these nodes out
only in virt-v2v's trimming function.

Thanks
Laszlo