Re: [PATCH 28/28] libxl: xsrestrict QEMU

[Ian Campbell <ian.campbell@citrix.com>]

On Tue, 2015-12-22 at 18:45 +0000, Ian Jackson wrote:
> If QEMU supports xsrestrict, pass xsrestrict=on to it (by default).
> 
> XXX We need to do this only if xenstored supports it, and AFAICT there
> is not a particularly easy way to test this.  Should we open a new
> test xenstore connection to query this information ?  We could do this
> once per libxl ctx.

Is this because the support is in oxenstored ^ cxenstored?

Otherwise I would argue that the toolstack should expect the xenstored to
be at least as new as it is.

> Let's start checking libxl.  The function that reads the physmap is
> libxl__toolstack_save.  It calls:
> 
> * libxl__xs_directory on /physmap
>   This is safe.

Potentially very many entries and hence a big return array? Are the guest
quotas sufficient to not worry about this?

> Either way libxl is fine.
> 
[...]
> Either way QEMU is fine.

Perhaps we should add a code comment to each of these places noting that
the values are guest controlled, since it is a bit more unexpected in this
case than e.g. frontend dirs.

> diff --git a/docs/misc/xenstore-paths.markdown b/docs/misc/xenstore-
> paths.markdown
> index 1dc54ac..27981de 100644
> --- a/docs/misc/xenstore-paths.markdown
> +++ b/docs/misc/xenstore-paths.markdown
> @@ -339,6 +339,13 @@ A PV console backend. Described in
> [console.txt](console.txt)
>  Information relating to device models running in the domain. $DOMID is
>  the target domain of the device model.
>  
> +#### ~/device-model/$DOMID/physmap [w]
> +
> +Information about an HVM domain's physical memory map.  This is
> +written as well as read by device models which may run at the same
> +privilege level of the guest domain.  When the device model ruus with

"runs"

> +system privilege, this path is INTERNAL.


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