4:39 a.m.
On 01/28/13 23:27, Eric Blake wrote:
On 01/11/2013 10:00 AM, Peter Krempa wrote:
> This patch documments XML elements used for (basic) support of virtual
s/documments/documents/
> RNG devices.
>
> In the devices section in the domain XML users may specify:
>
> <devices>
> <rng model='none'/>
Do we need model='none'? Historically, we have added it for devices
where the device used to be always-on by default, but we didn't have XML
for the device, then later on the device became optional so we had to
have a way to express the absence of the device without breaking
back-compat of old XML that gets the default for free. But in the case
of rng, I don't think we currently have any device for free, so
model='none' doesn't seem to add anything.
Hm, I was using the none model for testing purposes and decided to leave
it in. I found it useful to be able to disable the device without having
to delete the complete definition.
I agree that the none model is not strictly needed to be formatted in
the XML but I would still leave it as a option to be parsed.
> </devices>
>
> and the more useful variant:
>
> <devices>
> <rng model='virtio'>
> <source type='random'>/dev/urandom</source>
> </rng>
> </devices>
This part, however, seems reasonable.
> ---
> docs/formatdomain.html.in | 54 +++++++++++++++++++++++++++++++++++++++++++
> docs/schemas/domaincommon.rng | 31 +++++++++++++++++++++++++
> 2 files changed, 85 insertions(+)
>
> diff --git a/docs/formatdomain.html.in b/docs/formatdomain.html.in
> index bb0b199..7a5f267 100644
> --- a/docs/formatdomain.html.in
> +++ b/docs/formatdomain.html.in
> @@ -4260,6 +4260,60 @@ qemu-kvm -net nic,model=? /dev/null
> </ul>
> </dd>
> </dl>
> + <h4><a name="elementsRng">Random number generator
device</a></h4>
> +
> + <p>
> + The virtual random number generator device allows the host to pass
> + through entropy to guest operating systems.
> + <span class="since">Since 1.0.2</span>
Yikes - our delay in reviewing this means that it might not make 1.0.2
after all. It's probably a good sign that the list is now getting so
many patches, but at the same time it points out that everyone needs to
help more with reviews.
It's not that important for 1.0.2 anyways. I wanted to start a round of
reviews as I'm planing on adding other backends and stuff but I wanted
to do the initial design right.
> + <dt><code>model</code></dt>
> + <dd>
> + <p>
> + The required <code>model</code> attribute specifies what type
> + of RNG device is provided. Valid values are specific to
> + the virtualization platform:
> + </p>
> + <ul>
> + <li>'none' — disable the rng device</li>
Again, isn't omitting the <rng> element sufficient for this task?
It is.
> + <li>'virtio' — supported by qemu and virtio-rng
kernel module</li>
> + </ul>
> + </dd>
> + <dt><code>source</code></dt>
> + <dd>
> + <p>
> + The <code>source</code> element specifies the source of
entropy
> + to be used for the doimain. The source type is configured using the
s/doimain/domain/
> + <code>type</code> attribute. Supported source types are:
> + </p>
> + <ul>
> + <li>'none' — no source was configured</li>
> + <li>'random' — /dev/random or similar device as
source</li>
Here, it _does_ make sense to expose both a type='none' and
type='random'; assuming that qemu has a default source of entropy that
it uses when libvirt doesn't specify one.
> + </ul>
> + </dd>
> + <dt><code>source type='random'</code></dt>
> + <dd>
> + <p>
> + This source type expects a non-blocking character device as input.
> + Examples of such devices are /dev/random and /dev/urandom. The file
> + name is specified as contents of the <code>source</code>
element.
Side question - is it also possible to pass a (large) regular file, so
that you can experiment with reproducible random replays? At least with
coreutil's 'shuf --random-source', it is indeed possible to replay a
particular byte stream (obviously, on reply, the element of true
randomness is gone, but for analysis purposes, it can still be a useful
thing to do).
I haven't tried that configuration yet. It might be possible with the
current state of qemu and it will be possible when I implement the
entropy distribution daemon. There you will be able to configure a
source either from the kernel's interfaces, hardware generators or a
file for testing purposes.
I guess what this means for libvirt is that we should allow whatever
file name the user passes, rather than actually stat()ing the file and
enforcing that it be a char device.
I will try that and tweak this appropriately.
>
> + <define name="rng">
> + <element name="rng">
> + <attribute name="model">
> + <choice>
> + <value>none</value>
> + <value>virtio</value>
> + </choice>
Again, not sure we need this level of <choice>.
virtio model will be sufficient for most use cases.
> + </attribute>
> + </element>
> + </define>
> +
> + <define name="rng-source">
> + <element name="source">
> + <choice>
> + <group>
> + <attribute name="type">
> + <value>none</value>
> + </attribute>
> + <empty/>
> + </group>
> + <group>
> + <attribute name="type">
> + <value>random</value>
> + </attribute>
> + <ref name="filePath"/>
> + </group>
> + </choice>
> + </element>
> + </define>
This part looks fine.
Peter