Re: kern/53616: dom0 crashes when dumping domU images with vnd+fss

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Wed, Sep 19, 2018 at 08:52:51AM +0200, Frédéric Fauberteau wrote:
> Le 2018-09-19 07:59, Manuel Bouyer a écrit :
> > I don't think that will help. The problem is the fs is modified by the
> > VM
> > while dom0 has it mounted.
> 
> At the best of my understanding, xen uses vnd to mount image files:

no they are not mounted. The image files are attached to a vnd device,
but the filesystem is not mounted in the dom0

> # vndconfig -l
> vnd0: /srv/xen (/dev/mapper/vg0-lvxen) inode 3
> vnd1: /srv/xen (/dev/mapper/vg0-lvxen) inode 4
> vnd2: /srv/xen (/dev/mapper/vg0-lvxen) inode 5
> vnd3: /srv/xen (/dev/mapper/vg0-lvxen) inode 5
> 
> I have launched my backup process. vnd3 is the image file manually attached.
> But vnd{0,1,2} seem to have been mounted by xen (if I understand correctly).
> The filesystem of my third VM is mounted twice by dom0.
> Do you think a could make a fss snapshtop over vnd2 (third VM image) without
> manually attach it to vnd3 ?

You can't snapshot a vnd device alone in the dom0, if its content is used
in the domU. You have to shapshot the filesystem where the disk image
files are stored.

> 
> > The closest way to mimic what you could do with lvm would be to snapshot
> > the filesystem containing the disk images, and mount the snapshot to
> > backup
> > them.
> 
> Here, I could make a snapshot of /dev/mapper/vg0-lvxen. Thanks for the idea!
> 
> Do you think I have a chance to identify from where the problem exactly
> comes? Or it will be better to close this PR due to misuse of dom0?

It is clearly a misuse of dom0.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--