Re: [PATCH RFC/for-4.2?] libxl: Support backend domain ID for disks

[Ian Jackson <Ian.Jackson@eu.citrix.com>]

Ian Campbell writes ("Re: [PATCH RFC/for-4.2?] libxl: Support backend domain ID for disks"):
> On Mon, 2012-08-06 at 22:51 +0100, Daniel De Graaf wrote:
> > Allow specification of backend domains for disks, either in the config
> > file or via xl block-attach.
> > 
> > A version of this patch was submitted in October 2011 but was not
> > suitable at the time because libxl did not support the "script=" option
> > for disks in libxl. Now that this option exists, it is possible to
> > specify a backend domain without needing to duplicate the device tree of
> > domain providing the disk in the domain using libxl; just specify
> > script=/bin/true (or any more useful script) to prevent the block script
> > from running in the domain using libxl.

Thanks, Daniel!

> Given that this patch was originally posted so long ago, that the
> script= stuff just went in, that driver domains were on the TODO at one
> point (I think) and the relative simplicity of this patch I'm leaning
> towards taking this in 4.2.

The patch looks good to me and it improves a regression compared to
xend, the fixing of which is one of our goals for 4.2.  So I think
this is fine.

> I'm not sure if using libxl in libxlu is a layering violation or not
> (perhaps Ian J has an opinion), but I suppose it is acceptable (the
> alternative is a twisty maze of callbacks).

No, it's not a layering violation.

The only question in my mind was whether the libxl domain config
struct (in the IDL) should contain (perhaps optionally) the name, so
that the lookup is done at a slightly better time, but the result
would not be pretty at all and I think it's better this way.

> If we are going to expose libxl down to libxlu maybe we should go all
> the way and add the ctx to the XLU_Config?

Yes, I think that would be better.

Ian.