[RFC] Enhancement of loop driver?

[RFC] Enhancement of loop driver?

[Jürgen Groß]

[-- Attachment #1.1.1: Type: text/plain, Size: 1232 bytes --]

Hi,

a large customer is asking for storage migration of the disk images of
their virtual machines. They don't want to migrate the VM to another
host, but only the disk image from one filer to another while the VM
keeps running.

The natural way to setup something like this would be LVM and use
mirroring, but the problem is that this requires to copy the image to
a LVM enabled disk first, and this is no option due to time constraints
(copying many GB of data takes too long, and in the end I'd like to be
able to do the switch from the original image to the LVM backed with
the VM kept running).

So my idea was to enhance the loop driver to be capable to support a
list of backing files instead of only one and use a small prepended
file for storing the needed LVM metadata, resulting in the ability to
keep the existing disk image.

Would such an addition be acceptable?

An alternative would be to have an additional layer on top of the
current loop driver doing the concatenation of multiple loop devices,
but this would require a lot of code duplication. And using a common
base driver for the common code would be more code churn than just
adding the support to loop.c.

Thoughts?


Juergen

[-- Attachment #1.1.2: OpenPGP_0xB0DE9DD628BF132F.asc --]
[-- Type: application/pgp-keys, Size: 3135 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]

Re: [RFC] Enhancement of loop driver?

[Christoph Hellwig]

On Fri, Nov 13, 2020 at 07:59:23AM +0100, J??rgen Gro?? wrote:
> a large customer is asking for storage migration of the disk images of
> their virtual machines. They don't want to migrate the VM to another
> host, but only the disk image from one filer to another while the VM
> keeps running.
> 
> The natural way to setup something like this would be LVM and use
> mirroring, but the problem is that this requires to copy the image to
> a LVM enabled disk first, and this is no option due to time constraints
> (copying many GB of data takes too long, and in the end I'd like to be
> able to do the switch from the original image to the LVM backed with
> the VM kept running).
> 
> So my idea was to enhance the loop driver to be capable to support a
> list of backing files instead of only one and use a small prepended
> file for storing the needed LVM metadata, resulting in the ability to
> keep the existing disk image.
> 
> Would such an addition be acceptable?

Just use device mapper on top of the loop device, no need for changes
to the loop driver itself.  Also qemu has some interesting migration
features that you might want to look into as they might suit you.

Re: [RFC] Enhancement of loop driver?

[Jürgen Groß]

[-- Attachment #1.1.1: Type: text/plain, Size: 1448 bytes --]

On 13.11.20 09:44, Christoph Hellwig wrote:
> On Fri, Nov 13, 2020 at 07:59:23AM +0100, J??rgen Gro?? wrote:
>> a large customer is asking for storage migration of the disk images of
>> their virtual machines. They don't want to migrate the VM to another
>> host, but only the disk image from one filer to another while the VM
>> keeps running.
>>
>> The natural way to setup something like this would be LVM and use
>> mirroring, but the problem is that this requires to copy the image to
>> a LVM enabled disk first, and this is no option due to time constraints
>> (copying many GB of data takes too long, and in the end I'd like to be
>> able to do the switch from the original image to the LVM backed with
>> the VM kept running).
>>
>> So my idea was to enhance the loop driver to be capable to support a
>> list of backing files instead of only one and use a small prepended
>> file for storing the needed LVM metadata, resulting in the ability to
>> keep the existing disk image.
>>
>> Would such an addition be acceptable?
> 
> Just use device mapper on top of the loop device, no need for changes
> to the loop driver itself.  Also qemu has some interesting migration
> features that you might want to look into as they might suit you.
> 

Thanks. I just stumble over device mapper myself. :-)

I'd like to have a solution without qemu in order to be able to support
kernel based backends, too.


Juergen

[-- Attachment #1.1.2: OpenPGP_0xB0DE9DD628BF132F.asc --]
[-- Type: application/pgp-keys, Size: 3135 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]