Re: Guest Agent issue with 'guest-get-osinfo' command on Windows

["Richard W.M. Jones" <rjones@redhat.com>]

On Thu, Sep 02, 2021 at 02:36:51PM +0100, Daniel P. Berrangé wrote:
> On Thu, Sep 02, 2021 at 03:36:01PM +0300, Konstantin Kostiuk wrote:
> > Hi Team,
> > 
> > We have several bugs related to 'guest-get-osinfo' command in Windows Guest
> > Agent:
> > https://bugzilla.redhat.com/show_bug.cgi?id=1998919
> > https://bugzilla.redhat.com/show_bug.cgi?id=1972070
> > 
> > This command returns the following data:
> > {
> > "name": "Microsoft Windows",
> > "kernel-release": "20344",
> > "version": "N/A",
> > "variant": "server",
> > "pretty-name": "Windows Server 2022 Datacenter",
> > "version-id": "N/A",
> > "variant-id": "server",
> > "kernel-version": "10.0",
> > "machine": "x86_64",
> > "id": "mswindows"
> > }
> > 
> > The problem is with "version" and "pretty-name". Windows Server
> > 2016/2019/2022 and Windows 11 have the same MajorVersion ("kernel-version")
> 
> Yes, this is a long standing issue with version mapping Windows
> guests, to which no one has ever come up with a nice solution
> that I know of.
> 
> In libosinfo database, we just report the kernel version as the
> OS version, and accept the fact that there's a clash in version
> between various Windows products.
> 
>   https://gitlab.com/libosinfo/osinfo-db/-/blob/master/data/os/microsoft.com/win-2k19.xml.in
> 
>   https://gitlab.com/libosinfo/osinfo-db/-/blob/master/data/os/microsoft.com/win-10.xml.in
> 
> Apps that need to distinguish simply have to look at the
> product name, even if this causes localization pain.
> 
> Similarly in libguestfs, the virt-inspector tool just reports
> the kernel version, and product name from the registry:
> 
> # virt-inspector -d win2k8r2
> <?xml version="1.0"?>
> <operatingsystems>
>   <operatingsystem>
>     <root>/dev/sda2</root>
>     <name>windows</name>
>     <arch>x86_64</arch>
>     <distro>windows</distro>
>     <product_name>Windows Server 2008 R2 Standard</product_name>
>     <product_variant>Server</product_variant>
>     <major_version>6</major_version>
>     <minor_version>1</minor_version>
> 
> 
> # virt-inspector -d win10x64
> <?xml version="1.0"?>
> <operatingsystems>
>   <operatingsystem>
>     <root>/dev/sda2</root>
>     <name>windows</name>
>     <arch>x86_64</arch>
>     <distro>windows</distro>
>     <product_name>Windows 10 Pro</product_name>
>     <product_variant>Client</product_variant>
>     <major_version>10</major_version>
>     <minor_version>0</minor_version>
>     <windows_systemroot>/Windows</windows_systemroot>
>     <windows_current_control_set>ControlSet001</windows_current_control_set>
>     <hostname>DESKTOP-GR8HTR3</hostname>
>     <osinfo>win10</osinfo>

We actually try to turn it into a libosinfo compatible short string as
you can see from Dan's second example above and this code:

  https://github.com/libguestfs/libguestfs/blob/master/lib/inspect-osinfo.c

Which is I think what every tool should return.  libosinfo is the only
project that attempts to classify a broad range of OSes and is
constantly being updated.

> > This solution has several problems: need to update the conversion matrix
> > for each Windows build, one Windows name can have different build numbers.
> > For example, Windows Server 2022 (preview) build number is 20344, Windows
> > Server 2022 build number is 20348.
> > 
> > There are two possible solutions:
> > 1. Use build number range instead of one number. Known implementation
> > issue: Microsoft provides a table (
> > https://docs.microsoft.com/en-Us/windows-server/get-started/windows-server-release-info)
> > only with stable build numbers. So, we exactly don't know the build number
> > range.
> 
> Yep, this looks troublesome when considering non-GA releases.
> 
> > 2. We can read this string from the registry
> > (HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion). Known
> > implementation issues: ProductName value is localized (in a Russian version
> > of Windows, the word "Microsoft' is translated), so we should ignore it.
> > ReleaseId value does not equal to Windows Server version (for Windows
> > Server 2019, ReleaseId is 1809)
> 
> This reg key is what libguestfs reports IIUC
> 
> https://github.com/libguestfs/libguestfs/blob/master/daemon/inspect_fs_windows.ml#L227
> 
> > In conclusion, I have the next questions:
> > What solution we should implement to get the Windows release name?
> > Does someone know how end-users use this information? Should it be English
> > only or it can be localized? Should we have exactly the same output as now?
> > What should we do with the 'Standard' server edition? Currently, the guest
> > agent always returns 'Datacenter'.
> 
> This is equiv ot libguestfs' "product variant" data,w hich it gets
> from InstallationType registry key
> 
>   https://github.com/libguestfs/libguestfs/blob/master/daemon/inspect_fs_windows.ml#L259
> 
> Personally I think there's value in having consistent treatment of this
> info across qemu guest agent and libguestfs / libosinfo.

Agree.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html