Re: [Qemu-devel] PATCH: Control over drive open modes for backing file

[Anthony Liguori <anthony@codemonkey.ws>]

Jamie Lokier wrote:
> Anthony Liguori wrote:
>   
>> So while I think it's valid to have a "read-only disk" exposed to the 
>> guest, I don't think the user should have anything to do with how we 
>> open the file.
>>     
>
> All good points.
>
> My concern (and it may be different to others) is when I branch an
> image using a qcow2 with backing qcow (perhaps more than two deep).
>
> The backing image is sometimes shared with other branches.
>
> E.g. I might have a full install of an OS, and then have a branch
> where I install software package A, and another branch with software
> package B.  Both share the same base image.
>
> In these, it's very important that I don't accidentally modify the
> base image.  E.g. if I foolishly entered the 'commit' monitor command
> in VM A, I'd *corrupt* the disk of VM B.
>   

Introducing an option to keep you from misusing another option is a 
little silly.  You're just as likely to forget to use the first option 
as you are to "accidentally" issue a commit.

If you want to protect yourself, chmod the base image.  Then you have 
nothing to worry about.

> So it would be nice to open the base image read-only.
>
> (Microsoft have a similar description in their documentation on 'disk
> image differencing for maintaining test systems' for Virtual PC).
>
> Now savevm snapshots (as opposed to -snapshot) confuse this picture.
>
> Can I arbitrarily branch off different snapshots within a single qcow2
> file?  Can I delete the 'base' snapshots safely?  Most important: what
> happens if I have snapshots in a qcow2 file which has a base file, and
> I type 'commit'?  Does it corrupt all the snapshots, effectively?
>   

snapshots are linear, not branching.  And they are only every 
saved/loaded from the highest level qcow2 file.

Regards,

Anthony Liguori

> I find the snapshot facility a bit unclear as to what _exactly_ it
> does to be honest, and am therefore wary of it.
>
> Thanks,
> -- Jamie
>
>
>