missing block script support for qemu in libxl

missing block script support for qemu in libxl

[Olaf Hering]

Why does libxl now allow script= with backend=tap|qdisk? See
tools/libxl/libxl_device.c:disk_try_backend.

Ideally the script should prepare the backend storage based on info from
target=. Then the script should report either the dentry to be used by
qemu back to libxl, or it should setup the "physical-device" entry to be
used by blkback.

Right now the target= info is passed to the script, but its also used as
->pdev_path for qemu. Obvisouly qemu cant possibly make sense of the
target= info, because its meant for the script=.

tools/libxl/check-xl-disk-parse has some script examples. I think right
now its not possible to boot an HVM guest from iscsi, for example.

Should the logic in libxl be changed to make scripts= more useful, and
let libxl read info for qemu which has to be provided by the script?

Olaf

Re: missing block script support for qemu in libxl

[George Dunlap]

On Wed, Nov 18, 2015 at 9:25 AM, Olaf Hering <olaf@aepfle.de> wrote:
> Why does libxl now allow script= with backend=tap|qdisk? See
> tools/libxl/libxl_device.c:disk_try_backend.
>
> Ideally the script should prepare the backend storage based on info from
> target=. Then the script should report either the dentry to be used by
> qemu back to libxl, or it should setup the "physical-device" entry to be
> used by blkback.
>
> Right now the target= info is passed to the script, but its also used as
> ->pdev_path for qemu. Obvisouly qemu cant possibly make sense of the
> target= info, because its meant for the script=.
>
> tools/libxl/check-xl-disk-parse has some script examples. I think right
> now its not possible to boot an HVM guest from iscsi, for example.
>
> Should the logic in libxl be changed to make scripts= more useful, and
> let libxl read info for qemu which has to be provided by the script?

You mean, should the bug wherein HVM domains with emulated disks
(which is all of them, by default) cannot use block scripts be fixed?
Yes it should, and I'm working on it at the moment.

 -George

Re: missing block script support for qemu in libxl

[George Dunlap]

On Wed, Nov 18, 2015 at 9:35 AM, George Dunlap
<George.Dunlap@eu.citrix.com> wrote:
> On Wed, Nov 18, 2015 at 9:25 AM, Olaf Hering <olaf@aepfle.de> wrote:
>> Why does libxl now allow script= with backend=tap|qdisk? See
>> tools/libxl/libxl_device.c:disk_try_backend.
>>
>> Ideally the script should prepare the backend storage based on info from
>> target=. Then the script should report either the dentry to be used by
>> qemu back to libxl, or it should setup the "physical-device" entry to be
>> used by blkback.
>>
>> Right now the target= info is passed to the script, but its also used as
>> ->pdev_path for qemu. Obvisouly qemu cant possibly make sense of the
>> target= info, because its meant for the script=.
>>
>> tools/libxl/check-xl-disk-parse has some script examples. I think right
>> now its not possible to boot an HVM guest from iscsi, for example.
>>
>> Should the logic in libxl be changed to make scripts= more useful, and
>> let libxl read info for qemu which has to be provided by the script?
>
> You mean, should the bug wherein HVM domains with emulated disks
> (which is all of them, by default) cannot use block scripts be fixed?
> Yes it should, and I'm working on it at the moment.

Or, if you're keen, I could hand it off to you. :-)  But since I'm
currently elbow-deep in the domain creation code it's probably easier
for me to just press on.

 -George

Re: missing block script support for qemu in libxl

[Olaf Hering]

On Wed, Nov 18, George Dunlap wrote:

> You mean, should the bug wherein HVM domains with emulated disks
> (which is all of them, by default) cannot use block scripts be fixed?

Yes.

> Yes it should, and I'm working on it at the moment.

Great. Please cc me when the patches are ready.

Olaf

Re: missing block script support for qemu in libxl

[Paul Durrant]

> -----Original Message-----
> From: xen-devel-bounces@lists.xen.org [mailto:xen-devel-
> bounces@lists.xen.org] On Behalf Of Olaf Hering
> Sent: 18 November 2015 09:45
> To: George Dunlap
> Cc: xen-devel@lists.xen.org
> Subject: Re: [Xen-devel] missing block script support for qemu in libxl
> 
> On Wed, Nov 18, George Dunlap wrote:
> 
> > You mean, should the bug wherein HVM domains with emulated disks
> > (which is all of them, by default) cannot use block scripts be fixed?
> 
> Yes.
> 
> > Yes it should, and I'm working on it at the moment.
> 
> Great. Please cc me when the patches are ready.
> 

Just wondering if you guys are aware of this work:

http://xapi-project.github.io/xapi-storage/

(more detail at http://xapi-project.github.io/xapi/futures/smapiv3/smapiv3.html)

It looks like a pretty neat way to deal with different storage backends and incorporates provisioning 'plugins' aswell.

  Paul

> Olaf
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel