KVM/Qemu VMs with a multi-screen Spice console – II – local access with remote-viewer via a network port

In my first article in this series on local and remote connections to the Spice console of a KVM/Qemu virtual machine [VM] I have given you an overview over some major options. See the central drawing in

KVM/qemu with the graphical multi-screen Spice console – I – Overview over local and remote access methods

In this post we explore a part of the drawing's left side - namely a local connection with the remote-viewer client over a network-port and the "lo"-device. Local means that we access the Spice console with a Spice client in a desktop session on the KVM/Qemu host.

You can find some basic information on "remote-viewer" here.

Note: "remote-viewer" supports VNC, too, but we ignore this ability in this series.

A local scenario with remote-viewer on a KVM/Qemu host corresponds to desktop virtualization - though done with a remote tool. Such an approach can become a bit improper, because setting up ports for potential remote network connections always has security implications. This is one of the reasons why I actually want to discuss two different local access methods which can be realized with remote-viewer:

  • Method 1: Access via a network port (TCP socket). This standard method is well documented on the Internet both regarding the VM's setup (e.g. with virt-manager) and the usage of remote-viewer (see the man pages). The network based scenario is also depicted in my drawing.
  • Method 2: Access via a local Unix socket. This method is not documented so well - neither regarding the VM's setup (e.g. with virt-manager) and not at all by the man pages for remote-viewer.

You may complain now that I never indicated anything in the above drawing about a pure Unix socket! Well, at the time of writing the first article I did not even know that remote-viewer worked over a pure Unix socket and how one can configure the VM for a Unix socket. I will discuss the more common "Method 1" in this present article first. "Method 2" will be the subject of the next post - and then I will provide an update of the drawing. Talking about sockets: When you think a bit about it - Unix sockets must already be an ingredient of the local interaction of libvirt-tools with qemu, too, ....

Below I will discuss a set up for "Method 1" which is totally insecure. But it can be realized in a very simple way and is directly supported by virt-manager for the VM setup. Thus, it will give us a first working example without major efforts for creating TLS certificates. But, you are/were warned:

Disclaimer:
The test-configurations I discuss in this and the next articles of this series may lead to major security risks. Such configurations should NOT be used in productive multi-user and/or network environments without carefully crafted protection methods (outside the qemu-configuration) - as local firewall protection, system policies, etc. I take no responsibility whatever for an improper usage of the ideas presented in this article series.

Assumptions for test scenarios

  • The KVM/Qemu host in this article series is an Opensuse Leap 15.2 host. Most of the prescriptions discussed may work with some minor modifications also on other Linux distributions. But you must look up documentations and manuals ...
  • I assume that you already have set up a KVM/Qemu based test-VM with virsh or virt-manager. Let us call the VM "debianx". Actually, it is a Kali system in my case. (Most of the discussed measures will work on guests with a Debian OS, too.)
  • On our Opensuse KVM host we find a XML-definition file "debianx.xml" for a libvirt domain (i.e. a VM) below the directory "/etc/libvirt/qemu/", i.e. "/etc/libvirt/qemu/debianx.xml". It has a typical XML structure and can be edited directly if required.
  • The KVM host itself has a FQDN of "MySRV.mydomain.de". (MySRV is partiall writtenin big letters to remind you that you have to replace it by your own domain.
  • The user on the KVM host, who invokes remote-viewer to access the VM, has a name "uvma". We will later add him to a group "libvirt" which gets access rights to libvirt tools. We shall also work with a different user "uvmb" who is a member of the group "users", only. Both users work on a graphical desktop on the KVM/Qemu host (!) - let us say a KDE desktop. These users open one or more Spice client windows on the host's desktop to display graphical desktops of the KVM/Qemu guest VMs - as if the windows were screens of the VM.

We must carefully distinguish between the (KDE) desktop used by "uvma" on the KVM host and the desktop of the guest VM:

The Spice client windows on the KVM host's (or later on independent remote systems) desktop present full graphical desktops (Gnome, XFCE, KDE ...) of the VM and allow for interactive working with the VM. The desktop used on the host can, of course, be of a different type than the one used on the guest.

I will use this first practical article also

  • to have a brief look at two basic configuration files for libvirtd and qemu settings,
  • to discuss some basic setup options for a Qemu VM,
  • to check the multi-screen ability of remote-viewer
  • and to prove the "one-seat" situation for the Spice console.

I restrict my hints regarding the guest configuration to Kali/Debian or Opensuse Leap guest systems on the VM. I provide images for a Kali guest system, only. I do not have time for more ...

Make backups

Before you start any experiments you should make a backup of the directories "/etc/libvirt" and "/etc/pki" of your host - and in doubt also of the disk-files (or real partitions) of your existing VMs. You should also experiment with a non-productive test VM, only.

Basic security considerations

(In-) Security is always a bit relative. Let us consider briefly, what we intend to do with respect to "Method 1":

With the first method we define a network port for unencrypted connections without authentication.

In principle the port can be used by anyone having access to your network or anyone present on your PC. Such a configuration is therefore only reasonable for desktop virtualization

  • if you really are alone on your PC
  • AND if you have a local firewall preventing access to the port from any other system AND if there is a local policy restricting access to the Qemu-hypervisor or Spice-clients.

Otherwise it opens a much too big attack surface. Remember: Anyone being able to use the network port can kick you off your Spice console session seat, remotely, locally and also when you are already logged into the guest OS. And:

All data transfer over any kind of network device (lo, ethernet, virtual devices on the host, ...) occurs unencrypted. Other users on your machine (legal users, e.g. active via ssh, or attackers) may with some privilege escalation be able to read it.

Access to the Spice console itself can, however, be restricted to users who know a password. At the end of this article we will add such a simple password to our configuration to get at least some basic protection against kicking you out of a Spice console session.

Preparational step 1: Settings in "/etc/libvirt/qemu.conf" and the integration of Spice with Qemu

As mentioned in the last article, Spice is fully integrated with the Qemu-emulator. So, it will not surprise you that we can modify some general properties of connections to a VM's Spice console by parameter settings in a configuration file for Qemu. The relevant file in the libvirt-dominated virtualization environment of Opensuse Leap 15.2 is

"/etc/libvirt/qemu.conf".

Readers of my last article may ask now: Why do we need to configure the Qemu-driver of libvirt at all, when we access the Spice console with remote-viewer directly via Qemu ? I.e., without passing a libvirt-layer? As depicted in the drawing?

Answer: Probably you and me start KVM/Qemu based VMs through "virsh" or "virt-manager", i.e. libvirt components. Then libvirt drivers which are responsible to start the VMs according to its setting profile must know what we expect from the Qemu emulator. If you worked directly with the qemu-emulator (by calling "/usr/bin/qemu-x86_64") you would have to specify a lot of options to control the outcome! Some sections below you will find an example. Libvirt-programs only automate this process.

"qemu.conf" allows for general default settings of a variety of qemu-parameters. Some of them can, however, be overwritten by VM-specific settings; see the next section. When you scroll through the file you come to a major section which contains Spice parameters. For an easy beginning of our experiments we set some selected parameters to the following values:

 
#spice_listen = "0.0.0.0" # just leave it as it is 
# .... 
spice_tls = 0  # This corresponds to a deactivation of TLS and a INSECURE configuration!
spice_tls_x509_cert_dir = "/etc/pki/libvirt-spice"
# .... 
# just leave the next settings as it is - it creates a socket at  "/var/lib/libvirt/qemu/domain-1-debianx/monitor.sock"
#spice_auto_unix_socket = 1  
spice_sasl = 0

Our first setting deactivates TLS. This is reasonable as long a we have not created valid TLS server certificates and encryption keys. But we can already define a default directory into which we later place such a certificate and key files. We also deactivate any SASL authentication support for the time being. (Actually, even if you left it at the default value of "1" it would not help you much - as on Opensuse systems a required file "/etc/sasl2/qemu.conf" is missing, yet. As standard SASL encryption methods are insecure today, SASL activation makes sense in combination with TLS only.

General security precautions in "/etc/libvirt/qemu.conf"
Talking about security: On an Opensuse system with activated "apparmor" you could/should set the following parameters in "/etc/libvirt/qemu.conf":

 
security_driver = "apparmor"
security_default_confined = 1

to confine VMs.

Note: If "apparmor" is deactivated for some reason you will after these settings not be able to use virt-manager or virsh. You can check the status of apparmor via "rcapparmor status" or systemctl status apparmor".
(Off topic: Those interested in process separation should also have a look at cgroup-settings for the VMs. I would not change the basic settings, if you do not know exactly what you are doing. But folks trying to experiment with "virgl3D" may need to add "/dev/dri/renderD128" to the device list for ACLs.)

Preparational step A: A brief look a "/etc/libvirt/libvirtd.conf" - allow a selected standard user to use virt-manger or virsh

As we are within the folder "/etc/libvirt", we take the chance to very briefly look at another configuration file, which will become more interesting in later articles:

/etc/libvirt/libvirtd.conf

Among other things this file contains a variety of parameters which control the access to libvirt tools via Unix sockets or TCP-sockets - with and without TLS and/or some form of authentication. This is of no direct importance for what we presently are doing with remote-viewer - as we access Spice console directly via qemu, i.e. without a libvirt-layer. However, for convenience reasons you should nevertheless be able to use virt-manager or virsh. If you do not want to work as root all the time you, i.e. he normal user "uvma", need(s) special access rights to access Unix sockets which are automatically opened by libvirtd. How can this be achieved on an Opensuse Leap system?

You can configure a special user-group (e.g. a group named "libvirt") to get access to virt-manager. You can do this in a section named "UNIX socket access controls". The instructions in Opensuse's documentation on virtualization (see: Connections and authorization and section 9.1.1.1 therein) will help you with this. Afterwards you must of course add your selected and trusted user (here: "uvma") to the defined group.

Note: The libvirtd-process is nevertheless run by the user "root" and not with the rights of the standard user (here "uvma"). The group just allows for a kind of sudo execution. Note also that the central sockets by which libvirt-clients connect to the libvirtd-daemon are created by systemd.

By the way - who is the user for qemu-processes on an Opensuse Leap system?
On an Opensuse there is a special user "qemu" which is used to run the Qemu-emulator processes for VMs. We also find a related group "qemu". User and group have special rights regarding aes-keys for potential credential encryption related to a VM. The user and its group also define the access right to special Unix sockets generated directly by the Qemu emulator and not by libvirtd. We shall request such a Unix socket from Qemu in the next article.

On other operative systems you may find a different pre-configured group for libvirt access rights. In addition the standard qemu-user may have a different name. For Debian and derivatives you find related information at the following links:
whats-the-purpose-of-kvm-libvirt-and-libvirt-qemu-groups-in-linux,
install-configure-kvm-debian-10-buster,
libvirt_qemu_kvm_debian .

Preparational step B: Restart libvirtd or reload its configuration after changes to qemu.conf, livirtd.conf or changes to XML-definition files for VMs

Changes of the file "/etc/libvirt/qemu.conf" or "/etc/libvirt/libvird.conf" require that the libvirtd daemon is restarted or forced to reload its configuration. The same is true for any changes of XML-definition files for VMs located in "/etc/lbvirt/qemu/". (We will change such files later on). According to the man page sending a SIGHUP signal to the daemon will enforce a reload of the configuration. You can indirectly achieve this by the commands

rclibvirtd reload [on Opensuse systems)
or
systemctl reload libvirtd [any system with systemd]

But: Any libvirt or qemu changes will NOT affect already running VMs. You have to stop them, then restart libvirtd, reload the daemon's configuration and restart the VMs again.

Preparational step C: Configure the VM to use a Spice-console - and some other devices

Ok, let us prepare our VM for "Method 1". As we are in a libvirt environment we could use virt-manager to perform the required Spice configuration (see the image in the first article) . But we can also directly edit the XML configuration file "/etc/libvirt/qemu/debianx.xml" - assuming that you already have a working configuration for your VM. You need to be root to edit the files directly. See the restrictive access rights of the domain definition files

-rw------- 1 root root   4758 Mar  8 13:33 debianx.xml

By the way: A detailed description of possible parameter settings is provided at https://libvirt.org/formatdomain.html. Search for Spice and qxl .....

Regarding "Method 1" I use the following settings. The listing below only gives you an excerpt of the XML domain file, but it contains the particularly relevant settings for "graphics" (can be looked upon as a display device) and "video" (an be interpreted as a kind of virtual graphics card):

Excerpts from a libvirt domain definition XML file

   
   <interface type='network'>
      <mac address='52:54:00:93:38:4c'/>
      <source network='os'/>
      <model type='virtio'/>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x03' function='0x0'/>
    </interface>
   ...
   <graphics type='spice' port='20001' autoport='no' listen='0.0.0.0' keymap='de' >    
      <listen type='address' address='0.0.0.0'/>
      <image compression='off'/>
    </graphics>
   ... 
    <sound model='ich6'>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x04' function='0x0'/>
    </sound>
    ...
    <video>
      <model type='qxl' ram='262144' vram='65536' vram64='2097152' vgamem='65536' heads='4' primary='yes'/>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x02' function='0x0'/>
    </video>
...

Parameters which affect general qemu-related properties (as e.g. the "listen"-settings) will overwrite the settings in "qemu.conf". The choice of the PCI slot numbers may be different for your guest.

You see that I defined a network (!) port for the VM by which we can access the Spice console data stream directly. I have changed this port from the default "5900" to "20001".

You should be aware of the fact that this sets a real network port on the KVM host - i.e. a network stack is invoked when accessing the Spice console. As indicated above we can avoid the superfluous network operations by directly working with a Unix socket (Method 2), but this is the topic of the next forthcoming article.

Regarding network port definitions for a Spice console three rules apply:

  • Each of the VMs defined requires its own individual port !
  • Firewalls between the client and the VM have to be opened for this port. This includes local firewalls on the VM, on the KVM host (with rules that may apply to virtual bridges, on the client system and on any network components in between.
  • A TLS port has later to be configured in addition if you deactivated the "autoport" functionality - as we have done in the example above. I shall describe how to set the TLS port in a forthcoming article.)

The individual port definitions already imply the provision of a URI with port specification when starting remote-viewer (see below).

Let me briefly comment on the QXL-settings:

The QXL configuration shown above allows for 4 so called "heads" (or connectors) of the virtual graphics device. These heads support up to 4 virtual screens, if requested by the Spice-client. Note that the situation is different from real hardware consoles:
The (Spice) client (here remote-viewer) defines the number of screens attached to the VM's console by a number of Spice client windows opened on the user's desktop on the KVM host. The dimensions of these Spice client windows determine the dimensional capabilities of the virtual "screens".

How the virtual screen resolution is handled afterwards may depend on the user's own desktop resolution (on the KVM host or a remote system) AND the display settings of the VM's guest system's own tools for screen management. Today these are typically tools integrated into the guest's graphical desktop (KDE, Gnome, XFCE) itself. Present Gnome and KDE5 desktops on the VM's guest Linux system automatically adapt to the virtual display dimensions - i.e. the Spice window dimensions. But a XFCE desktop on the guest may behave differently; see below.

A short note also on the network settings for our test VM:
A virtual machine will normally be associated with a certain type of virtual network. At least if you want to use it for the provision of services or other network or Internet dependent tasks. You can configure virtual networks with virt-manager (menu "edit" >> "connection details"); but you must be root to do this:

In my case I prefer a separate virtual network for which we need routing on the KVM host. This allows us to configure restrictive firewall rules - e.g, with netfilter on the host. It even allows for the setup of virtual VLANs - a topic which I have extensively covered in other article series in this blog. And when required we just can stop routing on the host.

Guest requirements
Note that the QXL-configuration must also be supported on the VM guest itself: You need a qxl-video-driver there - and it is useful to also have the so called spice-vdagent active (see the my referenced series about a qxl-setup in the last article). On the guest system this requires

  • the installation of the "spice-vdagent" package and the activation of the "spice-vdagent.service" (both on Debian/Kali and Opensuse guest systems) ,
  • the installation of the packet "xserver-xorg-video-qxl" on a Debian/Kali-guest or "xf86-video-qxl" on an Opensuse guest

Present Debian, Kali or Opensuse operative system on the KVM/Qemu guest VM should afterward automatically recognize the QXL-video-device and load the correct driver (via systemd/udev).
Regarding the (virtual) network device on the VM itself you must use the guest system's tools for the configuration of NICs and of routes.

Preparational step D: Starting the VM - and "Where do I find logs for my VM"?

I assume that you as user "uvma" are logged in to a graphical desktop session on your KVM host "MySRV". If you are not yet a member of the libvirt-privileged group, you may need to use sudo for the next step. On a terminal window you then enter "virt-manager &", select our VM "debianx" and boot it.

If you look at the process list of your KVM host you may find something like:

qemu     17662  5.5  6.8 15477660 4486180 ?    Sl   18:11   2:29 /usr/bin/qemu-system-x86_64 -machine accel=kvm -name guest=debianx,debug-threads=on -S -object secret,id=masterKey0,format=raw,file=/var/lib/libvirt/qemu/domain-1-debianx/master-key.aes -machine pc-i440fx-2.9,accel=kvm,usb=off,vmport=off,dump-guest-core=off -cpu Skylake-Client -m 8192 -overcommit mem-lock=off -smp 3,sockets=3,cores=1,threads=1 -uuid 789498d2-e025-4f8e-b255-5a3ac0f9c965 -no-user-config -nodefaults -chardev socket,id=charmonitor,fd=33,server,nowait -mon chardev=charmonitor,id=monitor,mode=control -rtc base=utc,driftfix=slew -global kvm-pit.lost_tick_policy=delay -no-hpet -no-shutdown -global PIIX4_PM.disable_s3=1 -global PIIX4_PM.disable_s4=1 -boot strict=on -device ich9-usb-ehci1,id=usb,bus=pci.0,addr=0x5.0x7 -device ich9-usb-uhci1,masterbus=usb.0,firstport=0,bus=pci.0,multifunction=on,addr=0x5 -device ich9-usb-uhci2,masterbus=usb.0,firstport=2,bus=pci.0,addr=0x5.0x1 -device ich9-usb-uhci3,masterbus=usb.0,firstport=4,bus=pci.0,addr=0x5.0x2 -device virtio-serial-pci,id=virtio-serial0,bus=pci.0,addr=0x6 -blockdev {"driver":"file","filename":"/virt/debs/debianx.qcow2","node-name":"libvirt-2-storage","cache":{"direct":true,"no-flush":false},"auto-read-only":true,"discard":"unmap"} -blockdev {"node-name":"libvirt-2-format","read-only":false,"cache":{"direct":true,"no-flush":false},"driver":"qcow2","file":"libvirt-2-storage","backing":null} -device virtio-blk-pci,scsi=off,bus=pci.0,addr=0x7,drive=libvirt-2-format,id=virtio-disk0,bootindex=1,write-cache=on -device ide-cd,bus=ide.0,unit=0,id=ide0-0-0 -netdev tap,fd=35,id=hostnet0,vhost=on,vhostfd=36 -device virtio-net-pci,netdev=hostnet0,id=net0,mac=52:54:00:93:38:4c,bus=pci.0,addr=0x3 -chardev pty,id=charserial0 -device isa-serial,chardev=charserial0,id=serial0 -chardev socket,id=charchannel0,fd=37,server,nowait -device virtserialport,bus=virtio-serial0.0,nr=1,chardev=charchannel0,id=channel0,name=org.qemu.guest_agent.0 -chardev spicevmc,id=charchannel1,name=vdagent -device virtserialport,bus=virtio-serial0.0,nr=2,chardev=charchannel1,id=channel1,name=com.redhat.spice.0 -device usb-tablet,id=input0,bus=usb.0,port=1 -spice port=20001,addr=0.0.0.0,disable-ticketing,plaintext-channel=default,image-compression=off,seamless-migration=on -k de -device qxl-vga,id=video0,ram_size=268435456,vram_size=67108864,vram64_size_mb=2048,vgamem_mb=64,max_outputs=4,bus=pci.0,addr=0x2 -device intel-hda,id=sound0,bus=pci.0,addr=0x4 -device hda-duplex,id=sound0-codec0,bus=sound0.0,cad=0 -chardev spicevmc,id=charredir0,name=usbredir -device usb-redir,chardev=charredir0,id=redir0,bus=usb.0,port=2 -chardev spicevmc,id=charredir1,name=usbredir -device usb-redir,chardev=charredir1,id=redir1,bus=usb.0,port=3 -device virtio-balloon-pci,id=balloon0,bus=pci.0,addr=0x8 -object rng-random,id=objrng0,filename=/dev/urandom -device virtio-rng-pci,rng=objrng0,id=rng0,bus=pci.0,addr=0x9 -sandbox on,obsolete=deny,elevateprivileges=deny,spawn=deny,resourcecontrol=deny -msg timestamp=on

 
You see that the user "qemu" started the qemu-process by calling "/usr/bin/qemu-system-x86_64" and adding a bunch of parameters. You may identify some parameters which stem from the settings in libvirt's XML definition file for our special the VM (or "domain").

Afterwards, have a look at connections with netstat on the host; you should see that something (qemu) listens on the defined Spice port of the VM:

MySRV:/etc/libvirt/qemu # netstat -a -n | grep 20001
tcp        0      0 0.0.0.0:20001           0.0.0.0:*               LISTEN
...

Regarding logs of the VM for error analysis: You will find them in the directory " /var/log/libvirt/qemu" (on an Opensuse Leap system).

Method 1: Local access to the Spice console with remote-viewer

Now, we are prepared to test a local access to the Spice console with remote-viewer. (As we work via a network port you may still need to open the Spice port on a local firewall - it depends of how you configured your firewall.) The format of the URI we have to specify for remote-viewer is

spice://FQDN_OR_IP:PORT_NR

with

  • FQDN_OR_IP: You have to present a host address which can be resolved by DNS or a directly an IP-address.
  • PORT_NR: The port for the Spice service - here "20001".

For our setup this translates to

spice://localhost:20001

and thus the command

remote-viewer spice://localhost:20001

As user "uvma" we enter this command in another terminal window of our graphical desktop on the KVM host :

uvma@MySrv:~> remote-viewer spice://localhost:20001

(remote-viewer:22866): GSpice-WARNING **: 12:43:58.150: Warning no automount-inhibiting implementation available

Just ignore the warning. Then you should get a window with a graphical display of some login-screen on the VM. Details depend of course on the configuration of your VM's operative system. In my case, I get the "gdm3"-login-screen of my virtualized Kali system:

The desktop-manager may be a different one on your system. After a login to a Gnome desktop I use remote-viewer's menu to open a 2nd screen:

and get two resizeable screen-windows :

As expected! When we change the dimensions the desktop of the VM adapts to the new virtual screen sizes. Just try it out.

Will this work with a KDE desktop on the Kali guest, too? Yes, it does. KDE5 and Gnome3 can very well coexist on a Kali, Debian or Opensuse Leap15.2 (guest) system! For a Kali system you find information how to install multiple graphical desktop environments e.g. at the following links:
install-kali-linux,
kde-environment-configuring-in-kali,
kali switching-desktop-environments.

The next image shows the transparency effect of a moving terminal window on a KDE desktop of the same Kali VM - again with 2 Spice screens requested by remote-viewer:

Such effects, which can also be configured on Gnome, require an active compositor on the guest's desktop. At least when working locally on a KVM host I never got any problems with the performance of QXL/Spice and remote-viewer with an active compositor on the guest's Gnome, KDE, and XFCE-desktops (without OpenGL acceleration). This is a bit different on a remote system and using remote-viewer over a LAN connection; see the next articles.

KDE and Gnome desktops on the VM automatically adapt to the sizes of the virtual screens. XFCE does not ... XFCE is the present default desktop on Kali; so I installed after a distribution upgrade. Here some remote-viewer hints for XFCE fans on a VM with Kali/Debian and XFCE:

  • Set the number of screens in remote-viewer to 1. After installing XFCE with " sudo apt update && sudo apt install -y kali-desktop-xfce" and a reboot we can choose "XFCE-Session" on a button of the gdm3-screen (after having entered the username). It will the boot into a XFCE session.
  • Then with the setting tool you can configure fixed resolutions for the screen displays (see below).
  • Or: Just deactivate a virtual display, resize the respective window on your host's desktop and reactivate the virtual display again in your XFCE guest. XFCE then automatically recognizes the new screen size and adapts.
  • When logging out choose an option to save the desktop.

Unfortunately, XFCE has no seamless coexistence with KDE or Gome on Kali, yet. The following sequence of steps may fail:
1) Log out of a XFCE desktop-session, 2) choose a KDE or Gnome session afterwards /(this works without rebooting via the gdm menu), 3) log out of KDE/Gnome again and 4) try to restart a XFCE desktop.
On my Kali guest (on an Opensuse KVM host) I always have to reboot the guest to get a working XFCE session again, even if I did not change the screen dimensions during the KDE/Gnome session .. 🙁

Netstat reveals multiple connections corresponding to Spice channels

Using remote-viewer with a netowrk port is a really simple exercise. Most of the text above was spend on looking at the general virtualization environment. But, actually, the situation behind the scenes is not so simple. When we use netstat again, we get the following:

  
MySRV:/etc/libvirt/qemu # netstat -a -n | grep 20001
tcp        0      0 0.0.0.0:20001           0.0.0.0:*               LISTEN     
tcp        0      0 127.0.0.1:20001         127.0.0.1:46738         ESTABLISHED
tcp        0      0 127.0.0.1:20001         127.0.0.1:46750         ESTABLISHED
tcp        0      0 127.0.0.1:46734         127.0.0.1:20001         ESTABLISHED
tcp        0      0 127.0.0.1:46750         127.0.0.1:20001         ESTABLISHED
tcp        0      0 127.0.0.1:20001         127.0.0.1:46742         ESTABLISHED
tcp        0      0 127.0.0.1:46738         127.0.0.1:20001         ESTABLISHED
tcp        0      0 127.0.0.1:46740         127.0.0.1:20001         ESTABLISHED
tcp        0      0 127.0.0.1:46748         127.0.0.1:20001         ESTABLISHED
tcp        0      0 127.0.0.1:20001         127.0.0.1:46740         ESTABLISHED
tcp        0      0 127.0.0.1:20001         127.0.0.1:46734         ESTABLISHED
tcp        0      0 127.0.0.1:20001         127.0.0.1:46728         ESTABLISHED
tcp        0      0 127.0.0.1:20001         127.0.0.1:46744         ESTABLISHED
tcp        0      0 127.0.0.1:46728         127.0.0.1:20001         ESTABLISHED
tcp        0      0 127.0.0.1:46742         127.0.0.1:20001         ESTABLISHED
tcp        0      0 127.0.0.1:20001         127.0.0.1:46748         ESTABLISHED
tcp        0      0 127.0.0.1:46744         127.0.0.1:20001         ESTABLISHED

We find many, namely eight (8), active established connections. What do these connections correspond to? Well, they correspond to individual data exchange channels; I quote from
https://libvirt.org/formatdomain.html#video-devices:

"When SPICE has both a normal and TLS secured TCP port configured, it can be desirable to restrict what channels can be run on each port. This is achieved by adding one or more elements inside the main element and setting the mode attribute to either secure or insecure. Setting the mode attribute overrides the default value as set by the defaultMode attribute. (Note that specifying any as mode discards the entry as the channel would inherit the default mode anyways.) Valid channel names include main, display, inputs, cursor, playback, record (all since 0.8.6 ); smartcard ( since 0.8.8 ); and usbredir ( since 0.9.12 )."

See also: spice_for_newbies.pdf

Theoretically, one should be able to use remote-viewer itself to activate or deactivate channels by setting up a "target file" for the connection; see the man-pages for more details. But on my machines it does not work neither locally not remotely. 🙁 I did not look into the details of this problem, yet ....

Testing the "one seat situation" at a Spice console

Now its time to test what I claimed in the last article: The Spice console can be accessed only by one user at a time. Let us assume, that user "uvma" is a member of the libvirt-group. Let us in addition create another user "uvmb" which is member of the group "users" on the host, only. I assume that "uvma" has opened a graphical desktop session on the KVM host (MySRV). E.g. a KDE session.

Step 1: As user "uvma" start virt-manager and boot your VM "debianx". On a terminal window (A) use remote-viewer to open the Spice console of the VM with 2 screens. Log into a graphical desktop of the VM.

Terminal A:

uvma@MySRV:~> remote-viewer -v  spice://localhost:20001
Guest (null) has a spice display
Opening connection to display at spice://localhost:20001

(remote-viewer:20735): GSpice-WARNING **: 18:21:31.344: Warning no automount-inhibiting implementation available

Step 2: On your host-desktop open another terminal window (N) and login via "su - uvmb" as the other user. As "uvmb" now enter "remote-viewer spice://localhost:20001".

uvma@MySRV:~> su - uvmb
Passwort: 
uvmb@MySRV:~> remote-viewer -v  spice://localhost:20001

This will at once close the open two Spice windows and after a blink of a second open up one or two new Spice windows again. The effect is clearly visible.

On terminal A we get:

Guest debianx display has disconnected, shutting down
uvma@MySRV:~> 

On terminal B we get a lot of warnings - but still the spice windows open in exactly the same status as we left them.

uvmb@MySRV:~> remote-viewer -v  spice://localhost:20001

(remote-viewer:20903): dbind-WARNING **: 18:25:40.853: Couldn't register with accessibility bus: Did not receive a reply. Possible causes include: the remote application did not send a reply, the message bus security policy blocked the reply, the reply timeout expired, or the network connection was broken.
Guest (null) has a spice display
Opening connection to display at spice://localhost:20001

(remote-viewer:20903): GSpice-WARNING **: 18:25:40.936: PulseAudio context failed Verbindung verweigert

(remote-viewer:20903): GSpice-WARNING **: 18:25:40.936: pa_context_connect() failed: Verbindung verweigert
Cannot connect to server socket err = Datei oder Verzeichnis nicht gefunden
Cannot connect to server request channel
jack server is not running or cannot be started
JackShmReadWritePtr::~JackShmReadWritePtr - Init not done for -1, skipping unlock
JackShmReadWritePtr::~JackShmReadWritePtr - Init not done for -1, skipping unlock
AL lib: (EE) ALCplaybackAlsa_open: Could not open playback device 'default': Keine Berechtigung

(remote-viewer:20903): GLib-GObject-WARNING **: 18:25:40.962: g_object_get_is_valid_property: object class 'GstAutoAudioSink' has no property named 'volume'
AL lib: (EE) ALCplaybackAlsa_open: Could not open playback device 'default': Keine Berechtigung

(remote-viewer:20903): GLib-GObject-WARNING **: 18:25:40.964: g_object_get_is_valid_property: object class 'GstAutoAudioSrc' has no property named 'volume'
AL lib: (EE) ALCplaybackAlsa_open: Could not open playback device 'default': Keine Berechtigung
AL lib: (EE) ALCcaptureAlsa_open: Could not open capture device 'default': Keine Berechtigung

(remote-viewer:20903): GSpice-WARNING **: 18:25:41.015: record: ignoring volume change on audiosrc

(remote-viewer:20903): GSpice-WARNING **: 18:25:41.018: record: ignoring mute change on audiosrc

(remote-viewer:20903): GSpice-WARNING **: 18:25:41.021: Warning no automount-inhibiting implementation available

(remote-viewer:20903): GSpice-WARNING **: 18:25:41.052: playback: ignoring volume change on audiosink

(remote-viewer:20903): GSpice-WARNING **: 18:25:41.052: playback: ignoring mute change on audiosink

 
The warnings are mainly due to the fact that user "uvmb" has no rights to access the Pulseaudio context of the desktop of user "uvma". (I take care of Pulseaudio in one of the next articles ...) At the first switch between the users you may only get one Spice window as "uvmb". Choose two screens then. Afterwards you can switch multiple times between the users uvma and uvmb. Two windows get closed for the user with present Spice console session, two windows are opened for the requesting user, and so on.

Conclusion: "uvmb" and "uvma" can steal each other the Spice console with remote-viewer as and when they like.

By the way: A Spice console can be left regularly by a user by just closing the Spice client window(s). Note that this will not change the status of your desktop session on the host! This is left open - with all consequences.

Restricting access by setting a password

Let us prevent being thrown of the "seat" in front of our Spice console. As root change the <graphics> definition in VM's definition file to

...
 <graphics type='spice' port='20001' autoport='no' listen='0.0.0.0' keymap='de' passwd='spicemeandonlyme' defaultMode='insecure'>
      <listen type='address' address='0.0.0.0'/>
      <image compression='off'/>
      <gl enable='no'/>
    </graphics>
   ... 

If you haven't noticed: The relevant difference is the "passwd" attribute. You could have added this setting via virt-manager, too. Try it out.

Shut down and restart your VM. Any user that now wants to access the Spice console using "remote-viewer spice://localhost:20001" will be asked for a password. You yourself, too, if you close the console and try to reopen it.

Regarding security: The password transfer will to my understanding be AES encrypted. See Daniel P. Berrangé's blog https://www.berrange.com/posts/2016/04/01/improving-qemu-security-part-3-securely-passing-in-credentials/. But as long as the rest of the data exchange with the VM is not encrypted, this precaution measure won't help too much against ambitioned attackers.

Conclusion

Using remote-viewer locally on a KVM host in combination with a network port to access a VM's graphical Spice console is easy and convenient. It requires a minimum of settings. In addition it offers up to four virtual screens which you can open via a menu.
However, using a network port implies security risks. You should think about proper counter-measures first if you want to lean on "remote-viewer". Once configured in the way described above there is no libvirt-layer or anything else which may hinder ambitious attackers to analyze the unencrypted data flow between the Spice client and the VM through the host's network port.

In the next article

KVM/Qemu VMs with a multi-screen Spice console – III – local access with remote-viewer via a Unix socket

I want to show you how one can configure the VM such that remote-viewer can access the Spice console via a local Unix port. A network port is not opened in this approach. Security is none the less hampered a bit as we have to give a standard user some rights which normally only the "qemu" group has. But we shall take care of this aspect by using ACLs.

Links

Virtualization on Opensuse
https://doc.opensuse.org/ documentation/ leap/ virtualization/ single-html/ book.virt/index.html

Libvirtd daemon
https://www.systutorials.com/docs/linux/man/8-libvirtd/

Libvirt connection URIs
https://libvirt.org/uri.html

Find the Spice URI to use
https://access.redhat.com/ documentation/ en-us/ red_hat_enterprise_linux/7/ html/ virtualization_deployment_and_ administration_guide/ sect-Domain_Commands- Displaying_a_URI_ for_connection_to_a_graphical_display

TLS disabled
https://unix.stackexchange.com/ questions/ 148794/ how-to-create-kvm-guest-with-spice-graphics-but-tls-disabled-using-virt-install

General Kali installation
https://www.cyberpratibha.com/blog/how-to-update-and-upgrade-kali-linux-to-latest-version/

KDE, XFCE on Kali
https://technicalnavigator.in/kde-vs-xfce-which-one-is-better/

KVM/Qemu VMs with a multi-screen Spice console – I – Overview over local and remote access methods

KVM/Qemu based virtual machines [VMs] are a nice and comfortable part of my life with Linux. I use them both as dedicated entities for various services on a server host and as experimental test or work environments on my Linux workstation. Whereas the access to server VMs was dominated by SSH connections and CLI interaction so far, working with VMs on my PC was and is is characterized by graphical user-interfaces.

In the latter case we speak of desktop-virtualization: Connections to the virtual guests are set up locally via sockets (or via a network port over device "lo") on the very same PC or workstation which also serves as KVM host. Supported by a client/server model for graphics, KVM/Qemu provides multiple capable methods for the local display of the graphical output of a VM. And by graphical output I mean a desktop environment as KDE, XFCE or Gnome.

For KVM/Qemu the combination of the so called

  • "Spice display" and the Spice protocol
  • with drivers for potent virtual video devices (as QXL or Virtio)

offers a very convenient solution for the presentation of a VM's graphical desktop with high performance - which, in my experience, is considerably faster than most VNC variants.

In addition:
With the help of certain Spice clients one can access a local VM via a desktop display which extents over up to four "screens". All of these "screens" are displayed within scalable windows of your preferred basic desktop environment on your Linux machine - e.g. KDE, LXDE. This means that you can graphically work with a full fletched KDE desktop of the VM displayed in multiple windows within a KDE or Gnome environment on your local KVM/Qemu host (e.g your Linux workstation).

Can we get this comfort also with remote connections over a LAN? With encryption tunnels? The answer is yes - and related Qemu and Libvirt configurations are the main topics of this article series.

Why do I write about these subjects, at all? Recently, I needed to work with a KVM/Qemu guest remotely on an Opensuse Leap 15.2 server. I wanted to use a SSH tunneled connection with password authentication for a special graphical Spice client, namely "virt-viewer". Being a bit naive, I ran into some major trouble. This in turn forced me to look closer at different possibilities for remote connections to KVM/Qemu based VM. Unfortunately, I found Opensuse's documentation on Spice sparse - to say the best. But Spice gets especially interesting and somewhat complex with remote access methods, encryption and authorization.

In this article series I discuss some of the multiple configuration options I have experimented with. I restrict myself to Spice based solutions with the clients "virt-viewer", "remote-viewer" and also "virt-manager". In combination with SSH, TLS. I hope my experiences help other Opensuse Leap users to set up convenient remote solutions.

After giving some documentation links, I start with an overview over some selected remote access scenarios which are available. They are the most important ones and are all based on network ports. But, in the coming posts I will cover local access scenarios based on Unix sockets, too. And their combination with SSH.

Limitations

Unfortunately, I have to reduce expectations of readers who are out for "multi-user remote desktop solutions". This is not what the present Spice solutions with the named clients will offer you. See below.

A topic which I will not address in the forthcoming articles is the setup of a local VM with OpenGL acceleration on the graphics device of the KVM host. With Nvidia cards and drivers this is unfortunately a mess - and it is Nvidia's fault, in my opinion. Very similar, by the way, to Nvidia's EGL-based solution attempt for Wayland. All of this because of their own OpenGL handling and their declined support for GBM. So, let us not go there to avoid frustration ...

Basic preparations of a KVM/Qemu VM: Documentation and guides?

I assume that you are already familiar with the basic set up and configuration of a KVM/Qemu based virtual machine - e.g. with virsh or virt-manager. If not, you will find valuable hints for Opensuse Leap systems in SuSE's "Virtualization Guide"

However, as said above, Opensuse's documentation on virtualization does not give you much information about Spice. We only get a glimpse of it at the sections on "virsh" and "Qemu". The information is rather spurious there; one doesn't get a broader context. You may get a better understanding by reading introductory pages on the site "https://www.spice-space.org/", especially the user manual
https://www.spice-space.org/spice-user-manual.html.
Regarding special configuration settings of a VM, see https://libvirt.org/formatdomain.html.

Spice with multiple screens requires a non default configuration of the VM's QXL device

For a basic Spice setup there is not too much to do; "just" choose Spice for the display-settings of your VM and configure the VM's QXL graphics device. Such a device corresponds to something like a virtual graphics card. Well, it sounds easy, but when I tried it myself, I had to experiment. To help others a bit: I have already discussed in depth how we can configure a KVM/Qemu guest and its QXL device such that it supports multiple scalable high-resolution screens for Spice clients. See the following articles:

KVM/qemu mit QXL – hohe Auflösungen und virtuelle Monitore im Gastsystem definieren und nutzen – I
KVM/qemu mit QXL – hohe Auflösungen und virtuelle Monitore im Gastsystem definieren und nutzen – II
KVM/qemu mit QXL – hohe Auflösungen und virtuelle Monitore im Gastsystem definieren und nutzen – III
KVM/qemu mit QXL – hohe Auflösungen und virtuelle Monitore im Gastsystem definieren und nutzen – IV

These posts were written in German; but with a little bit of goodwill one can understand the configuration settings and related commands.

An alternative today is to use a "virtio" video device in the VM's configuration. I shall come back to this point in a later article.

What I did not cover in my previous posts on Spice was the broader spectrum of access options to a KVM/Qemu-based VM with libvirt/Space clients - both regarding local and remote connections. I only discussed how to use the "remote-viewer" client locally - despite the fact that its name already indicates something else. What I also disregarded in my articles in 2017 is the fact that there are various (remote) access methods - direct ones and some through a libvirt-layer on the KVM-host. As always with Linux there are multiple ways to achieve your objectives :-).

However, when we use Spice instead of VNC we speak about a "one person" or a "one seat" solution.

Spice and libvirt => multiple methods to access exactly one graphical console of a KVM/Qemu based VM

What you should be aware of when you think of using Spice for a graphical access to a KVM/Qemu VM is the following:

1. The Spice protocol and the Spice display for KVM/QEMU based VMs do NOT provide a multi-user service for graphical remote terminals or remote desktop displays - at least not in their default configuration.

According to what I have read, anything in this direction still seems to be experimental. Instead:

2. We access a data stream describing the graphical contents of one or multiple screens of a console which is virtually "attached" to the VM.

Let us call this device a multi-screen Spice console of the VM.

The Spice console solution is fully integrated into the QEMU-emulator. The console screen data and related sound can, however, also be displayed and played on a remote Linux system. And here we face a point which makes the user's life both convenient and complex at the same time :

3. The Spice console of a VM can be accessed by various protocols or protocol combinations. The Spice console is fully integrated with Qemu and directly accessible there via a (network) socket, but it is in addition accessible via libvirt components. This leads to a variety of connectivity options.

These aspects can be underestimated: Remote access for multiple screen requires sufficient band-width of the network. The second point excludes the application of Spice for multi-user scenarios. The third point opens up a range of local and remote access possibilities - some are more complicated to configure, some are more secure than others. Some options are deactivated in a default libvirt installation (not only on Opensuse Leap systems).

The most important connection options for standard libvirt/spice clients are summarized in the following schematic drawing.

The Spice console of a KVM/Qemu VM

The graphics shows a Linux KVM host at its center. KVM supports the VM emulator QEMU in the Linux kernel - thus making QEMU fast. Two virtual machines VM1 and VM2 are shown. VM1's configuration includes a Spice (console) display - e.g. set up according to the following excerpt of a configuration example for a VM (done with virt-manager):

Note that the configuration item is called "display" in virt-manager (and "graphics" in the XML-configuration of a VM). I think that in the case of Spice "Console" would have been a better name. I should also warn you that the displayed configuration may lead to security risks - if not accompanied by protective measures.

Note also that "virt-manager" does not allow for a detailed QXL-configuration required for multiple screens on the displayed dialog. But you can edit the XML-files of the VM for this purpose via the second tab (or an editor; you find the XML-files for VMs in the directory "/etc/libvirt/qemu" on Opensuse Leap systems):

The QXL configuration displayed here is an example - you may choose other parameters. (Or test out a "virtio" device; see a later article).

"One seat" console
In the lower right part of the first sketch above I have tried to express that a graphical Spice console corresponds to a "one console with one seat" situation. The console is only available for exactly one (authorized) user at a time. Note:

Any user having sufficient access rights may any time claim the console and throw you off the seat in front of the graphical console screen, if you do not secure the access to the console by a password (or other measures)!

We did not do this in the configuration example above; see the empty field for the password. I quote from the documentation:

"Spice does not support multiple connections to the same QEMU instance by default. So anybody who will connect to the same host and port can simply take over your session. You can solve this problem by using ticketing." (see: https://www.spice-space.org/spice-user-manual.html)

"Ticketing" means setting a password to access the VMs console screen; but still anyone knowing the password could claim the console - a password is no real solution for a multi-user situation. To say it clearly:

If your objective is to establish something like a multi-user graphical access to a KVM/QEMU-based virtual machine with a full desktop display for each user (i.e. a "remote desktop solution"), then think about something like a headless graphical server configuration of the KVM guest. On a Linux VM, you can achieve this by either implementing a X2GO server, a commercial NoMachine server, a VNC server-variant or even XRDP. Some of these solutions can easily be combined with SSH.
And do not forget: In a LAN with sufficient bandwidth "ssh -X" may already serve most practical purposes.

But in many cases a local or remote "one seat situation" with a full graphical desktop may be sufficient. Now, let us look at the other parts of the drawing.

Different access options to the Spice console of a KVM/Qemu-based VM

There is a multitude of options to connect to the Spice console (https://www.spice-space.org/spice-for-newbies.html). We shall describe them briefly in this first article. In forthcoming articles we shall have a closer look at each of them.

Option 1: Simple connection with "remote-viewer" via a Spice (network) port

The Qemu-emulator is able to handle requests over the Spice protocol on its own, i.e. without the invocation of special libvirt-support. The Spice solution offers connections to the console either on the basis of a pure Unix socket (local solution) or a network port (remote solution).

A network oriented solution, which can be realized with the Spice client "remote-viewer" is depicted on the left side of our central sketch:

You access the Spice console of a specific VM directly via a specific network port. Spice ports have to be configured independently for each of the VMs. I.e., if you have 2 VMs whose Spice consoles you want to access independently then you need to define two different ports for the Spice consoles. Actually, you won't be able to start a VM which has the same port associated with its Spice console as another already running VM.

Properties of a remote-viewer connection over a network port:

  • Default port? The standard Spice port is 5900 (which is also the port used by VNC). In the example displayed above I instead used port 20001. (Do not forget to open your firewalls for your chosen ports!)
  • Security by encryption? The connection to the port is not secured by any encryption. However, it is possible to define an additional specific TLS-secured port. We shall cover spice + TLS protocol combinations in another article.
  • User authentication on the KVM server host? Can optionally be achieved via SASL.
  • Preferred Spice client? The Spice client which uses the Spice protocol directly and without libvirt is "remote-viewer".
  • Works locally and/or remotely? Both.

Some additional hints:

In some old articles or posts on the Internet you may find descriptions of "virt-viewer" together with URIs like "spice://host:5900". This does not work any more - at least on my systems.

By setting proper attributes to the VM, we can in addition enforce that it only accepts secured connections. For those who are eager to experiment already: Where do we learn something about such machine specific settings in XML-format? The answer is:
https://libvirt.org/formatdomain.html

Option 2: Connection with "virt-viewer" or "virt-manager" via libvirtd and a defined TCP port

Besides "remote-viewer" there is another Spice client called "virt-viewer". This client is of special use when you work on a KVM host with multiple VMs. What some admins do not know is that it can be used via a simple socket with a specific TCP-port. You do not need SSH or TLS, if you can disregard security. This corresponds pretty much to what we got with remote-viewer and a Spice port. The difference is that we access the socket via the mediation of libvirtd components:

virt-viewer requires URIs which specify the target hypervisor for the libvirtd daemon - in our case "qemu". Such URIs differ from URIs used for remote-viewer as they always contain a two fold protocol definition.

We shall apply such URIs in a later article. In addition this access method requires a special (non-default) configuration of the libvirt-daemon. Among other things, you, of course, have to define the port. The ability to listen to an insecure TCP port is normally deactivated in the libvirtd configuration; we have to activate it manually. This is something we shall have a closer look at.

Unencrypted access to the Spice console via a special libvirtd port may be a useful solution for KVM guests sitting in a secured physical LAN-segment or behind a virtual bridge/firewall in a virtualized LAN-segment of the KVM host.

Properties of a virt-viewer connection via a TCP port

  • Default port? 16509.
  • Security by encryption? The connection to the port is not secured by any encryption. However, we have other options for virt-viewer to connect to the VM via SSH or TLS secured connections. See below.
  • Preferred Spice client?: virt-viewer (You cannot provide a valid URI to remote-viewer this way). However, this connection option also works with "virt-manager" (and its included graphical access to the Spice console).
  • User authentication? Can be achieved via SASL.
  • Works locally and/or remotely? Both.

on the KVM host.
Note that the access to libvirtd must be allowed for the user invoking the connection! This can be configured on Opensuse Leap systems by some policy and/or libvirtd settings.

Option 3: Connection with "virt-viewer" or "virt-manager" via SSH and libvirtd

A very convenient method to get a secure connection with Linux is to use SSH encryption. Therefore, we are not astonished to find that "virt-viewer" can be combined with SSH. This connection method also requires a special URI-format. And again: access to libvirtd must be allowed for the user using this type of connection.

Properties of a virt-viewer connection via a SHH port:

  • Default port? 22 or otherwise the defined SSH-port on your system.
  • Security by encryption? SSH.
  • Preferred Spice client?: virt-viewer (You cannot provide a valid URI to remote-viewer this way). However, this connection option also works with "virt-manager" (and its included graphical access to the Spice console).
  • User authentication? Via SSH (preferably public-key authentication)
  • Works locally and/or remotely? Both.

Both options 2 and 3 work with virt-manager, too. However, virt-manager does not provide you with a multi-screen access to the Spice console.

Note that a combination of SSH and remote-viewer is also possible; but then you would open a "SSH -X" connection first and run remote-viewer on the KVM/Qemu host and not on the network client system. I will present such a solution in a later article, too.

Conclusion

The Spice console of KVM/Qemu offers access to a graphical desktop of a virtual machine guest. Spice clients as "remote-viewer", "virt-viewer" and "virt-manager" can be used locally and remotely. The first two clients offer multiple screens - but all clients provide a one seat solution, only. This may, however, be sufficient for many use cases. There is a variety of local and remote access methods. Connections can be secured by TLS or SSH encryption.

In the next article

KVM/Qemu VMs with a multi-screen Spice console – II – local access with remote-viewer via a network port

I shall have a closer look at a local connection from the "remote-viewer" client to the Spice console of a VM.

Links

Spice console and Spice protocol
https://linuxhint.com/configure_spice_server_debian_10/
https://www.spice-space.org/spice-user-manual.html
https://www.spice-space.org/spice-for-newbies.html

Activate spice
https://www.linux-kvm.org/page/SPICE
https://linux-blog.anracom.com/2017/08/15/kvmqemu-mit-qxl-hohe-aufloesungen-und-virtuelle-monitore-im-gastsystem-definieren-und-nutzen-iv/
https://octetz.com/docs/2020/2020-05-06-linux-hypervisor-setup/

Nvidia 🙁
https://www.phoronix.com/scan.php?page=news_item&px=XDC2016-Device-Memory-API

Further articles in this series

KVM/Qemu VMs with a multi-screen Spice console – VIII – VM and user specific restrictions for remote-viewer connections – iptables and sudo
KVM/Qemu VMs with a multi-screen Spice console – VII – remote-viewer, qemu and SASL authentication
KVM/Qemu VMs with a multi-screen Spice console – VI – remote access with remote-viewer and TLS encryption
KVM/Qemu VMs with a multi-screen Spice console – V – remote access via remote-viewer, a network port and a SSH-tunnel
KVM/Qemu VMs with a multi-screen Spice console – IV – remote access via SSH, remote-viewer and a Unix socket
KVM/Qemu VMs with a multi-screen Spice console – III – local access with remote-viewer via a Unix socket
KVM/Qemu VMs with a multi-screen Spice console – II – local access with remote-viewer via a network port

Upgrade workstation from Opensuse Leap 15.1 to Leap 15.2

Two weeks ago I upgraded my main Linux-Workstation from Leap 15.1 to Leap 15.2. Some first impressions:

  • Leap 15.2 with KDE/X11 works relatively well. Some minor setting-options for the Plasma desktop have changed.
  • KDE/Wayland with Nvidia cards does not work - unusable for production.
  • No problems with QEMU based KVM- and LXC-container virtualization - my old Debian and Kali guest systems worked as expected on the Leap 15.2-host. No problems with multi-screen configurations of the guests with remote-viewer either.
  • Unexpected high power consumption of the Nvidia card - it has increased by about 5% to 8%. I am not sure whether this is a KDE or a driver problem.
  • The usual suspects for problems during the upgrade are the Nvidia drivers, Pulseaudio, VMware and this time - to my big surprise - CUPS.
  • You may have to change/reset some KDE config and rc-files in your home-directory - especially for the plasma desktop itself.
  • Chromium showed some glitches.

I shortly describe the upgrade process and some obstacles I had to overcome.

Basic upgrade process

I always prepare for a quick restoration of my running system configuration (here: Leap 15.1). Ahead of any upgrade activity, I therefore create a backup of my Leap 15.1 "/"-filesystem on an external drive and in addition create another copy in a target partition on one of my bootable disks. I use the "dd"-command for both purposes. As all productive application and development data as mails, project documents, PHP- and Python modules are on independent partitions or on network locations anyway, the risk of running into major problems afterwards is relatively small. In addition I comment certain critical entries in the "/etc/fstab" out.
In case of a major disaster a recovery can be achieved very quickly by just re-copying the copy of the Leap 15.1 partition back to its original place.

Regarding the backup of the "/"-filesystem to an external disk you have the choice between creating a copy of a partition and creating a (zipped) file. See e.g.:
https://www.poftut.com/linux-dd-command-backup-examples/
https://linoxide.com/linux-command/linux-dd-command-create-1gb-file/
I just made simple partition copies. Regarding the blocksize "bs" option of "dd" I am very generous with SSDs; I use a size like bs=1M or bs=4M to get a reasonable performance. When copying around 100 GB you probably do not want to wait for an hour :-). E.g.:

dd if=/dev/sdg1 of=/dev/sdh5 bs=4M status=progress

Warning 1:

If you copy your partition with "/"-filesystem into another partition, then the target partition must at least have the size of the original one. Really take care of this point! I always try to achieve the exact same size with a partition tool (gparted or YaST partitioner).

Warning 2:

Do not forget to change the UUID of the partition's copy if the target partition resides on an internal disk; you should never have two filesystems with the same UUID on a system! But write down the original UUID at a safe place. Use "uuidgen" and the "tune2fs -U" commands to change the copied filesystem's UUID, afterwards.

See: how-to-change-filesystem-uuid-2-same-uuid

And: In case of an emergency restoration, i.e. after having copied the backup filesystem into its old position, you, of course, have to change the UUID back to the old one.

Why are these UUID-precautions required?
Because of your boot-loader - probably Grub2. Depending on its configuration (and some other system settings) it will probably have created entries in the "/boot/grub2/grub.cfg" which refer to the UUIDs of the bootable partitions. And if such a UUID-reference cannot be resolved in a unique way you may in the best case boot a system you did not want to boot, but you may also experience worse conflicts. Now, you may ask:

What is the copy on the same or another internal hard disk good for anyway?
I create the copy on my main SSD (or another internal disk) not only for restoration purposes, but also for the purpose of transforming it into a fully bootable filesystem at its new position later on - i.e. after the upgrade. I shall return this point in an extra section below.

Upgrade process

After the backup procedures I followed the guideline of the Brasilian Linux Kamarada for the upgrade; see
https://kamarada.github.io/en/2020/09/01/linux-kamarada-and-opensuse-leap-how-to-upgrade-from-151-to-152

Their way of upgrading Opnesuse systems is straight-forward and well-tested. I also liked their recent inclusion of the "releasever"-variable ...

I tend to use multiple repositories besides the standard upgrade repository of Opensuse. Therefore, I had to delete quite an amount of repositories following the Kamarada recipe. If you are in the same position, you should make a screenshot of the repository-configuration in YaST to be able to reconfigure them after the upgrade. During the upgrade zypper may recognize many packages for which it has to change the repository. It is a bit of an ordeal to answer to all of the related questions, but I recommend to follow the questions with focused concentration: The order of offered alternatives may change!

Hint: Take care of sufficient filesystem-space! The download of new packages requires extra space during the upgrade period - and usually also the total size of used disk space after an upgrade typically tends to grow somewhat. Therefore: Have at least 20GB available on your "/"-filesystem. (LVM is your friend - if you use(d) it ...).

After the upgrade you have to reboot (init 6). If your graphics card is from Nvidia you probably will find yourself on some console terminal and not at a desktop-manager-login (e.g. sddm or gdm3) at the end of the boot process. We must install the (new) Nvidia drivers afterwards.

Nvidia re-setup

There are at least two ways of installing Nvidia-drivers on an Opensuse-system: 1) Directly, i.e. by a manual installation with the help of of the downloaded driver installation scripts. 2) With the help of YaST and the Nvidia community repository. I have used the latter option during the last 2 years. Either way you probably prohibited loading the Opensource Nouveau driver already at some point in the past - e.g. by blacklisting the related kernel module in some of the files in "/etc/modprobe.d". If not - it is time to do so now.

When you rebooted after the Leap 15.2 upgrade the old Nvidia modules did not work as they were not compiled for the new kernel which was installed during the upgrade. So, from the console login, which you hopefully have reached, you need to invoke YaST in ASCII-mode, activate the Nvidia community repository and reinstall the proprietary Nvidia drivers from there. In my case the packages

nvidia-computeG05, nvidia-gfxG05-kmp-default, nvidia-glG05, x11-video-nvidiaG05

A compilation is done automatically during this update.
Hint: I also suggest to issue "mkinitrd" immediately after the driver installation to be on the safe side regarding the "initramfs"-phase of the boot-process.
This worked pretty well in may case. After a reboot I got to my SDDM-login screen afterwards. My standard X11-based graphical KDE Plasma desktop started smoothly afterwards. I also tested the functionality of my Cuda installation for Keras and Tensorflow2 - worked perfectly.

Increased idle power consumption of the graphics card on a running KDE plasma desktop (Version 5.18)
A unwelcome surprise regarding Nvidia was the following: When watching the power consumption of my graphics card with

watch -n0.1 nvidia-smi

I saw a rise in the average consumption.
An average value of the GPU load (with "Force Composition Pipeline" activated on all my 3 screens) now had a value of 15%; on my old Leap 15.1 it was below 10%. All with having set the Powermizer mode to "Adaptive" (via (nvidia-settings). I fiddled a bit around with some Nvidia settings - but found no real solution, yet. When I changed a desktop theme (with the help of "systemsettings5") I observed a drop of power consumption to 11% - but after some movement of windows across my 3 screens it settles again at 15% - quite independent of style and desktop settings. At least with an OpenGL-compositor activated (systemsettings5 >> "screen ... >> compositor". Choosing "Xrender" leads to a drop of GPU-consumption to 9%.

What brought me down a bit to 12% with an activated OpenGL-compositor (OpenGL2 or OpenGL3.1) was to deactivate keeping window-previews:

I did not experience any real disadvantages by doing so. On Leap 15.1 the average consumption with a driver a bit older was still lower by 4% to 5%. We talk about 2 Watt here, but still I do not like such unexpected changes. I do not know whose fault this is - KDE's or Nvidia's. We shall see what happens with the next driver version ....

Reconfigure KDE settings

The screen order on my Plasma 5.18 desktop was OK. No screen-tearing (see: Opensuse, KDE Plasma, X11, Nvidia – stop video and screen tearing); my old "xorg.conf"-settings were respected and active.

However, when trying to change design- and plasma-style-settings via "systemsettings5" a click on the "plasma-style"-icon led to a window crash and a Plasma error. This is usually a sign that something is wrong with the KDE configuration files:

Because the KDE setting- and configuration files in the home-directory survive the upgrade untouched some may not be compatible with the new KDE version.

I do not know about an easy way to find out which of the files are to be recreated. My standard way is to make a copy of the "~/.config-directory (with a new suitable name), delete the original directory, wait until it is automatically created again (with new default settings). We make a copy of this new default-".config" version (with a unique name), too, and then copy the original version into its standard place again. In the end we have 3 ".config"- directories with different names - one in place with the user's original settings, one with the new default settings and one with the original settings as ".config".
Then I overwrite files in the ".config"-file with the new default files, at least those for which I suspect some problems - starting with the plasma-specific "plasma.....rc"-files, then the "systemsettingrc", then "kwinrc". Thus, the problems with "plasma-style"-settings could be resolved quickly.

I also got some problems with a standard activation of the OpenGL-compositor at the start of the plasma desktop. A change of the respective settings in the "kwinrc"-file helped - especially "Enabled=True":

[Compositing]
AnimationSpeed[$d]
Backend=OpenGL
Enabled=true
GLCore=true
GLPreferBufferSwap=a
GLTextureFilter=1
HiddenPreviews=4
OpenGLIsUnsafe=false
WindowsBlockCompositing=false
XRenderSmoothScale=false

Pulseaudio [PA]

A major frustration came up when I tried to start Clementine and listen to some songs - all my PA-settings were gone with the exception of the PA-LADSPA-equalizer. This had partially to do with changes in the Phonon-settings - now to be found under "systemsettings5 >> Audio". I had to turn off my onboard HD sound card, activate my mainly used Xonar D2X explicitly and also deactivate an existing XFI-Card, too.

The dialog for setting the KDE/Phonon "standard device" have changed once again :-(. You cannot choose priorities for different sources (audio, video, ...) any longer. Simplification? I hate this new KDE politics of taking more and more configuration options away from the standard user. I had to make the equalizer the "standard device" in the new dialog to direct all streams (from browsers and players) through this device.

Compare this to the settings in Leap 15.1; see KDE, Pulseaudio and Browsers – make the LADSPA equalizer the default sink.
In addition: We cannot play test sounds for the various sinks any more. This does no longer work in "YaST Audio" either ... I do not call such things progress.

With PA activated "pavucontrol" serves as my primary mixer. It worked more or less as expected though some functionality has been removed there, too. E.g. the ability to remove some input channels completely. Another stupid thing which happens now regarding the default channel for volume adjustment via kmix is now that you may choose "Simultaneous output ...." - but volume controls will affect the Xonar channels, too - at least as long as no device uses the equalizer ....

It took some fiddling until I got back most of my old functionality and sound quality. I found again that its worth to reduce the Clementine input in pavucontrol down to 80% to avoid some oversteering and sound-distortion. Do not forget to set the default channel for sound control with kmix to "Simultaneous output ..." for your multi-channel soundcard!

One can in addition install and activate the XFCE-mixer, which basically is a graphical frontend to an alsa mixer. This allows additional shifts of the different channels against each other - performed after pulseaudio's own mixer settings, namely those of "pavucontrol".

A major problem with sound and the onboard card:
For some reason a switch from the graphical terminal to a console terminal (by "Ctrl Alt F3") automatically activates an Nvidia HDMI stereo device on my system now. It comes up again even if I explicitly deactivated it in the KDE or PA audio settings. I have no solution for this, yet. It affects, however, at least a running Chromium (see below) - and leads to a restart of desktop effects when I return to the graphical session. I get a corresponding explicit message from Plasma ....

VMware - a yearly game of money ...

VMware Workstation 15.5 Pro does not work on a Leap 15.2 host without special prerequisites. See e.g.:
https://forums.opensuse.org/showthread.php/540779-vmware-worksation-pro-15-5-not-failed
https://www.opensuse-forum.de/thread/63542-vmware-workstation-pro-leap-15-2b/?pageNo=1
https://communities.vmware.com/t5/VMware-Workstation-Pro/Can-t-compile-wmware-workstation-15-5-on-leap-15-2/m-p/2286237

However, WS Pro 16.0 does. I upgraded to this version. My Win 10 guests felt a bit more sluggish than with WS 15.5. But this may also have to do with a subsequent Upgrade of Win 10 to the 2004 version.

By the way: Note that VMware wants you to pay extra maintenance support, if you order it from SW resellers - at least here in Germany. The cheapest choice was to order it directly from the VMware shop. Do we like such anti-competition monopoly and customer control measures? No, we do not like it => time to get Windows 10 guests running on KVM machines ... and to forget VMware for the future. See:
https://getlabsdone.com/10-easy-steps-to-install-windows-10-on-linux-kvm/#Customizing-the-Hardware-for-Windows-10

Hint: Be a bit reluctant with changing the HW compatibility of your VMware virtual machine for two reasons: You may want the virtual image to continue running in your old 15.1 implementation, too - until you feel that Leap 15.2 is stable enough. Second reason: Changing too much of HW settings may lead to a reactivation of your Windows license key.

CUPS bug - reconfigure printers - use the CUPS packages from the printing repository

I have two printers in my network. They can be addressed both by a local configuration and via a cups server in the net, which offers a variety of queues.
The reconfiguration of the local settings were a piece of cake with YaST. The updated HP-drivers simply worked.
However, printing from my system via a CUPS server in the net and its queues did not work any more. In contrast to all previous upgrades of at least the last 4 years :-(.
It took me a while to find out that this is due to a major bug in the present CUPS version delivered by the standard Update repository of Opensuse. Use the CUPS version > 2.3.0 from the "Printing repository".
Then your server based queues will work again!

Chromium: Problems with HW acceleration / sound from Chromium does not get upmixed on a multichannel sound card

I had some problems with Chromium. Maybe I had them already before, but I noticed them now:
I had to refrain from HW-acceleration: Whenever I changed to one of the console terminals (e.g. Ctrl-Alt-F2) and returned to the graphical terminal the Chromium user interface got totally destroyed. Not funny when you were editing a blog :-(.
In addition: Chromium sound does not get upmixed completely - the probable reason is that Chromium supports real 7.1 channel sound these days. So, stereo may remain stereo or only gets upmixed by some special soundcard setting to 4.0, 6.0 - but not by PA. Despite such settings on my Xonar card, I had some disturbing experiences: When changing to a console terminal and back to the graphical one, the upmixing changed from 4.0 to 2.0 .... :-(. When I rebooted afterwards, the 4.0 upmixing was there again. The whole effect probably was due to an automatic activation of some onboard/oncard Nvidia HDMI sound channels during the switch to the console terminals.

Firefox sound, however, which sends a pure stereo signal to the system as, gets properly upmixed by PA to 7.1 sound and this remains stable during switches between console and graphical terminals. So, for the time being, I do not use chromium for TV streaming any longer. Maybe the Chromium developers should have a talk with the PA developer .... It seems to be a difficult problem to me - you would have to analyze that only stereo sound is coming despite a 8 channel signal. A simple workaround would be that we get a switch in the chromium browser which allows to set its operation from multichannel to stereo, only.

For those who like to experiment with "pactl" commands (like "pactl list sinks" as standard user) I would like to draw your attention to work others ave done regarding PA and upmixing:
https://gist.github.com/dex4er/8646669
https://bbs.archlinux.de/viewtopic.php?id=16093

It should be possible to write two scripts one activating a certain upmixing with pactl commands and one deactivating it again. But my time is so limited that I haven't tried it myself, yet.

KDE Plasma, Nvidia and Wayland do not work

As with Leap 15.1, KDE and Wayland lead to a direct crash on my workstation with Nividia graphics. The system freezes. You have to enforce a reset of your PC and start a reboot. However, on a laptop with a i915 Intel driver for a CPU integrated graphics Wayland works. So, we have a clear Nvidia topic here.

After the upgrade: Make the copy of our original partition of Leap 15.1 a bootable one

If you had copied your original Leap 15.1 partition into another partition on your hard disk you may want to make it a bootable fully functional system offered as an entry in the Grub2 menu. This requires some additional efforts.

Firstly, you have to change all references to the modified UUID in the internal boot-relevant files of the moved filesystem - i.e. in the "/etc/fstab"  AND  in the "/boot/grub2/grub.cfg" there on the Leap 15.1-partition. For this purpose you must mount the copied Leap 15.1 partition on your running Leap 15.2 system and edit the entries in both files of the mounted Leap 15.1 filesystem very carefully with respect to the changed UUID (see the sections on backups above). Take care that you do not change the files on the running Leap 15.2 system. And, please, make copies of both files before you change them! You may need the original contents.

Instead ofchnaging entries in the "/boot/grub2/grub.cfg" of the Leap 15.1 copy you can also rename the "/boot/grub2/grub.conf" on the Leap 15.1 partition to something like "grub.conf_orig" (you want to keep the contents!). This gives the grub "os-prober" on you Leap 15.2 installation a chance to create correct entries regarding the address of the partition.

Then we remove the "/boot/grub2/grub.cfg" in the Leap 15.2 installation and run "grub2-mkconfig -o /boot/grub2/grub.cfg" again to give the os-prober there a chance to create valid entries. Please, cou check these entries first before you install the grub loader from your running Leap 15.2 again! If you find correct UUID or other references to the partition the you can reinstall the bootloader. I always do this with YaST. Afterwards you should be able to boot both your new Leap 15.2 and your old 15.1 installation a respective entries in the Grub2 menu. Thus we have our old Leap 15.1 installation available for comparisons.

Conclusion

An upgrade of a Linux workstation from Opensuse Leap 15.1 to Leap 15.2 brought no major problems with it. One remarkable exception is a bug in the CUPS package - for which you find a solution in form of a newer package in another Suse repository. The fact that Wayland does not work with KDE on Nvidia graphics came not unexpected - although I think its a shame for Nvidia. The same could be said about the increased power consumption of the Nvidia card.

Have fun with Leap 15.2.