Richard and Cedric, thank you for your replies.
On 07/06/2016 09:11 PM, Richard W.M. Jones wrote:
On Wed, Jul 06, 2016 at 05:33:01PM +0300, Maxim Perevedentsev wrote:
> Hello everyone!
>
> I have an issue with OpenSUSE-42.1 (I guess this is the only distro
> with btrfs snapshots by default).
> As we remember, list_filesystems returns btrfs snapshots along with devices.
>
>> <fs> list-filesystems
> /dev/sda1: swap
> btrfsvol:/dev/sda2/@: btrfs
> btrfsvol:/dev/sda2/@/.snapshots: btrfs
> btrfsvol:/dev/sda2/@/.snapshots/1/snapshot: btrfs
> btrfsvol:/dev/sda2/@/.snapshots/2/snapshot: btrfs
> ...
> btrfsvol:/dev/sda2/@/.snapshots/16/snapshot: btrfs
> /dev/sda2: btrfs
>
>
> The btrfs snapshots are not block devices, so an error is thrown by
> RESOLVE_DEVICE macro in stubs.c. That means we will get an error
> trying to call blockdev - oriented functions at these devices.
>
> =========================
>
> After grepping "list_filesystems" issues were found in:
> - virt-sparsify (in-place and copying): is_lv
I'll send a patch soon.
> - virt-filesystems: blockdev_getsize64
> - virt-cat (inspect_os) treats this as multiboot system and refuses
> to proceed
> - virt-rescue --suggest suggests all snapshots (maybe this is correct)
> - virt-sysprep --operation fs-uuids goes into infinite loop
>
> [ 144.2] Performing "fs-uuids" ...
> virt-sysprep: warning: cannot set random UUID on filesystem
> btrfsvol:/dev/sda2/@/.snapshots/16/snapshot type btrfs: set_uuid:
> set_uuid_stub: btrfsvol:/dev/sda2/@/.snapshots/16/snapshot: expecting a
> device name
The specific problem is that set_uuid is declared to take a Device as
the first parameter [see generator/action.ml], whereas I believe it
should take a Mountable instead.
Compare it to how vfs_uuid is declared in the same file.
This is simply a bug in generator/action.ml, and I believe a simple
one line patch should fix it.
I tried that, but this does not work.
btrfstune takes an unmounted filesystem as an argument.
It is impossible to change uuid for a subvolume. Passing a Mountable
results in setting different uuids to /dev/sda2 16 times.
I think it is better to skip subvolumes on this operation.
> virt-sysprep: warning: cannot set random UUID on filesystem
> btrfsvol:/dev/sda2/@/.snapshots/1/snapshot type btrfs: set_uuid:
> set_uuid_stub: btrfsvol:/dev/sda2/@/.snapshots/1/snapshot: expecting a
> device name
> virt-sysprep: warning: cannot set random UUID on filesystem
> btrfsvol:/dev/sda2/@/.snapshots type btrfs: set_uuid: set_uuid_stub:
> btrfsvol:/dev/sda2/@/.snapshots: expecting a device name
> virt-sysprep: warning: cannot set random UUID on filesystem
> btrfsvol:/dev/sda2/@ type btrfs: set_uuid: set_uuid_stub:
> btrfsvol:/dev/sda2/@: expecting a device name
> [ 154.4] Performing "fs-uuids" ...
>
> =========================
>
> I'd like to ask for your comments on fixing this.
> Should we
> 1. add a new command (guestfs_list_filesystems_devices) to return
> only device-like filesystems?
> 2. add something like guestfs_is_device which handles subvolumes on
> host side?
> NB: guestfs_is_blockdev also fails on stub
> 3. allow guestfs_is_(blockdev|chardev|...) to check subvolumes on
> host side? (maybe together with 2?)
> 4. anything else?
Number 4.
Rich.
Do I understand right that the right way of skipping btrfs subvolumes is
guestfs_mountable_device(fs) != NULL && guestfs_mountable_subvolume(fs)
!= NULL.
And we should add this code to every place where we need Device?
--
Your sincerely,
Maxim Perevedentsev