[libvirt] libxl and non-absolute paths

Wed Feb 18 13:23:12 UTC 2015

On 18.02.2015 04:15, Jim Fehlig wrote:
> Stefan Bader wrote:
>> Just recently we moved to libvirt 1.2.12 for the next release. Which brought up
>> a few problems when working with configs which we and Debian used to have.
>>
>> A mild complaint towards the xml validation: it would be really nice of that
>> would be a bit more specific about what exactly it complains. It took me a while
>> to realize that "Extra element os in interleave" was trying to tell me
>> that the string of the loader element within the os section was not an absolute
>> path.
>>
>> The issue here is that with libxl, I think the goal was to rather allow the
>> library to select the path prefix (like for pygrub where the full path got
>> removed recently). But now the xml validation disagrees.
>>
>> This would go for bootloader for xenpv and loader (within os) for xenfv. And for
>> emulator in the device section. Though for that things are a bit more
>> complicated. The libxl driver now calls that with the help option and decides
>> from the output whether this is the "traditional" xen forked qemu or the
>> upstream qemu binary. Then it selects the device model depending on that outcome.
>>   
> 
> It only sets the device_model_version
> (LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN vs
> LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN_TRADITIONAL).  device_model is
> directly VIR_STRDUP'ed from def->emulator.

Ok, I did not word that correctly. The main caveat here is that to figure out
the device_model_version it calls out to what is set in def->emulator. So it
needs a full path. Up to libvirt 1.2.8 at least the contents of def->emulator
were actually rather ignored since we ended up with "qemu-dm" which was not true
at all.

> 
>> Not sure whether the libxl driver could query libxl for the path prefix.
> 
> I asked about that too, but the Xen community preferred a build-time
> approach.  Wei Liu added a pkgconfig module to Xen, enabling build-time
> detection of the paths
> 
> commit babeca328413baebfdca366a5b17c06acf4295e8
> Author: Wei Liu <wei.liu2 at citrix.com>
> Date:   Fri Jan 9 14:32:18 2015 +0000
> 
>     libxl: provide xenlight.pc
>    
>     A pkg-config file for libxl. It also contains two variables
>     (xenfirmwaredir and libexec_bin) so that tools that are very keen on
>     knowing the locations of Xen binaries (say, libvirt) can use them to
>     determine the location of the binaries.
> 
> We are not yet making use of xenlight.pc in libvirt.  But this work aims
> to improve reporting the correct paths in *capabilities*. It could also
> be used in the xl.cfg to domXML conversion code.

Hm, ok. I guess there is upstream and upstream who might also be doing Debian
(and derivatives of that). Since libxl is part of a version package the
pathnames change between Xen versions. Maybe something to discuss about but not
really here. Reality now causes the full path requirement to become a major pain
in the butt.

> 
>> Right
>> now the most straight forward way seems to move back to a full path for the
>> emulator.
> 
> Full paths are required as per the documentation.  From
> http://libvirt.org/formatdomain.html#elementsOSBootloader
> 
> "The content of the |bootloader| element provides a fully qualified path
> to the bootloader executable in the host OS."
> 
> Same for <emulator>.
> 
>>  At least now, by using the standard qemu binary for everything, we got
>> a predictable path that does not change with Xen versions. So its possible to
>> force migrate over to put /usr/bin/qemu-system-i386 there.
>>
>> But for loader and bootloader, do you think it reasonable to change the
>> templates from absFilePath to filePath?
>>   
> 
> There's probably a fair bit of existing config with e.g.
> <bootlaoder>pygrub</bootloader> that no longer validates.  ATM, I don't
> have a good answer :-/.  Is correct capabilities information and xl.cfg
> -> libvirt.xml conversion, along with better error reporting enough? 
> Should users change their non-validating domXML?

Potentially this has to be magically converged in postparse of libxl. This is
where I currently move from the completely lying "qemu-dm" to
"/usr/lib/qemu-system-i386".

Though this will depend on using xenlight.pc and that depends on it being there.
The change currently only is on the devel master branch. So presumably Xen 4.6
may have it. For the reality of "now" I guess the best way around the issues is
to either disable validation completely or at least make that
absFilePath->filePath conversion. None of that though is worthy of being upstream.

-Stefan
> 
>>From libvirt's perspective, I think full paths, discovered from
> capabilities, should be required.  I'd like to hear Daniel's opinion. 
> He may have considered such cases when enabling the validation.
> 
> Regards,
> Jim
> 
>> -Stefan
>>
>> --- libvirt-1.2.12.orig/docs/schemas/domaincommon.rng   2015-01-23 12:46:24.0000
>> +++ libvirt-1.2.12/docs/schemas/domaincommon.rng        2015-02-13 10:00:43.1616
>> @@ -258,7 +258,7 @@
>>                  </choice>
>>                </attribute>
>>              </optional>
>> -            <ref name="absFilePath"/>
>> +            <ref name="filePath"/>
>>   
>>            </element>
>>          </optional>
>>          <optional>
>> @@ -1060,7 +1060,7 @@
>>        <optional>
>>          <element name="bootloader">
>>            <choice>
>> -            <ref name="absFilePath"/>
>> +            <ref name="filePath"/>
>>              <empty/>
>>            </choice>
>>          </element>
>>
>>   
> 
> --
> libvir-list mailing list
> libvir-list at redhat.com
> https://www.redhat.com/mailman/listinfo/libvir-list
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20150218/d10b134d/attachment-0001.sig>