QEMU (short form for Quick Emulator) is an open source hypervisor that emulates a physical computer. From the perspective of the host system where QEMU is running, QEMU is a user program which has access to a number of local resources like partitions, files, network cards which are then passed to an emulated computer which sees them as if they were real devices.
A guest operating system running in the emulated computer accesses these devices, and runs as if it were running on real hardware. For instance, you can pass an ISO image as a parameter to QEMU, and the OS running in the emulated computer will see a real CD-ROM inserted into a CD drive.
QEMU can emulate a great variety of hardware from ARM to Sparc, but {pve} is only concerned with 32 and 64 bits PC clone emulation, since it represents the overwhelming majority of server hardware. The emulation of PC clones is also one of the fastest due to the availability of processor extensions which greatly speed up QEMU when the emulated architecture is the same as the host architecture.
|
Note
|
You may sometimes encounter the term KVM (Kernel-based Virtual Machine). It means that QEMU is running with the support of the virtualization processor extensions, via the Linux KVM module. In the context of {pve} QEMU and KVM can be used interchangeably, as QEMU in {pve} will always try to load the KVM module. |
QEMU inside {pve} runs as a root process, since this is required to access block and PCI devices.
The PC hardware emulated by QEMU includes a motherboard, network controllers,
SCSI, IDE and SATA controllers, serial ports (the complete list can be seen in
the kvm(1) man page) all of them emulated in software. All these devices
are the exact software equivalent of existing hardware devices, and if the OS
running in the guest has the proper drivers it will use the devices as if it
were running on real hardware. This allows QEMU to run unmodified operating
systems.
This however has a performance cost, as running in software what was meant to run in hardware involves a lot of extra work for the host CPU. To mitigate this, QEMU can present to the guest operating system paravirtualized devices, where the guest OS recognizes it is running inside QEMU and cooperates with the hypervisor.
QEMU relies on the virtio virtualization standard, and is thus able to present paravirtualized virtio devices, which includes a paravirtualized generic disk controller, a paravirtualized network card, a paravirtualized serial port, a paravirtualized SCSI controller, etc …
|
Tip
|
It is highly recommended to use the virtio devices whenever you can, as
they provide a big performance improvement and are generally better maintained.
Using the virtio generic disk controller versus an emulated IDE controller will
double the sequential write throughput, as measured with bonnie++(8). Using
the virtio network interface can deliver up to three times the throughput of an
emulated Intel E1000 network card, as measured with iperf(1). [1]
|
Generally speaking {pve} tries to choose sane defaults for virtual machines (VM). Make sure you understand the meaning of the settings you change, as it could incur a performance slowdown, or putting your data at risk.
General settings of a VM include
-
the Node : the physical server on which the VM will run
-
the VM ID: a unique number in this {pve} installation used to identify your VM
-
Name: a free form text string you can use to describe the VM
-
Resource Pool: a logical group of VMs
When creating a virtual machine (VM), setting the proper Operating System(OS) allows {pve} to optimize some low level parameters. For instance Windows OS expect the BIOS clock to use the local time, while Unix based OS expect the BIOS clock to have the UTC time.
On VM creation you can change some basic system components of the new VM. You can specify which display type you want to use.
Additionally, the SCSI controller can be changed. If you plan to install the QEMU Guest Agent, or if your selected ISO image already ships and installs it automatically, you may want to tick the QEMU Agent box, which lets {pve} know that it can use its features to show some more information, and complete some actions (for example, shutdown or snapshots) more intelligently.
{pve} allows to boot VMs with different firmware and machine types, namely SeaBIOS and OVMF. In most cases you want to switch from the default SeaBIOS to OVMF only if you plan to use PCIe passthrough.
A VM’s Machine Type defines the hardware layout of the VM’s virtual motherboard. You can choose between the default Intel 440FX or the Q35 chipset, which also provides a virtual PCIe bus, and thus may be desired if you want to pass through PCIe hardware. Additionally, you can select a vIOMMU implementation.
Each machine type is versioned in QEMU and a given QEMU binary supports many machine versions. New versions might bring support for new features, fixes or general improvements. However, they also change properties of the virtual hardware. To avoid sudden changes from the guest’s perspective and ensure compatibility of the VM state, live-migration and snapshots with RAM will keep using the same machine version in the new QEMU instance.
For Windows guests, the machine version is pinned during creation, because Windows is sensitive to changes in the virtual hardware - even between cold boots. For example, the enumeration of network devices might be different with different machine versions. Other OSes like Linux can usually deal with such changes just fine. For those, the Latest machine version is used by default. This means that after a fresh start, the newest machine version supported by the QEMU binary is used (e.g. the newest machine version QEMU 8.1 supports is version 8.1 for each machine type).
Very old machine versions might become deprecated in QEMU. For example, this is
the case for versions 1.4 to 1.7 for the i440fx machine type. It is expected
that support for these machine versions will be dropped at some point. If you
see a deprecation warning, you should change the machine version to a newer one.
Be sure to have a working backup first and be prepared for changes to how the
guest sees hardware. In some scenarios, re-installing certain drivers might be
required. You should also check for snapshots with RAM that were taken with
these machine versions (i.e. the runningmachine configuration entry).
Unfortunately, there is no way to change the machine version of a snapshot, so
you’d need to load the snapshot to salvage any data from it.
QEMU can emulate a number of storage controllers:
|
Tip
|
It is highly recommended to use the VirtIO SCSI or VirtIO Block controller for performance reasons and because they are better maintained. |
-
the IDE controller, has a design which goes back to the 1984 PC/AT disk controller. Even if this controller has been superseded by recent designs, each and every OS you can think of has support for it, making it a great choice if you want to run an OS released before 2003. You can connect up to 4 devices on this controller.
-
the SATA (Serial ATA) controller, dating from 2003, has a more modern design, allowing higher throughput and a greater number of devices to be connected. You can connect up to 6 devices on this controller.
-
the SCSI controller, designed in 1985, is commonly found on server grade hardware, and can connect up to 14 storage devices. {pve} emulates by default a LSI 53C895A controller.
A SCSI controller of type VirtIO SCSI single and enabling the IO Thread setting for the attached disks is recommended if you aim for performance. This is the default for newly created Linux VMs since {pve} 7.3. Each disk will have its own VirtIO SCSI controller, and QEMU will handle the disks IO in a dedicated thread. Linux distributions have support for this controller since 2012, and FreeBSD since 2014. For Windows OSes, you need to provide an extra ISO containing the drivers during the installation.
-
The VirtIO Block controller, often just called VirtIO or virtio-blk, is an older type of paravirtualized controller. It has been superseded by the VirtIO SCSI Controller, in terms of features.
On each controller you attach a number of emulated hard disks, which are backed by a file or a block device residing in the configured storage. The choice of a storage type will determine the format of the hard disk image. Storages which present block devices (LVM, ZFS, Ceph) will require the raw disk image format, whereas files based storages (Ext4, NFS, CIFS, GlusterFS) will let you to choose either the raw disk image format or the QEMU image format.
-
the QEMU image format is a copy on write format which allows snapshots, and thin provisioning of the disk image.
-
the raw disk image is a bit-to-bit image of a hard disk, similar to what you would get when executing the
ddcommand on a block device in Linux. This format does not support thin provisioning or snapshots by itself, requiring cooperation from the storage layer for these tasks. It may, however, be up to 10% faster than the QEMU image format. [2] -
the VMware image format only makes sense if you intend to import/export the disk image to other hypervisors.
Setting the Cache mode of the hard drive will impact how the host system will notify the guest systems of block write completions. The No cache default means that the guest system will be notified that a write is complete when each block reaches the physical storage write queue, ignoring the host page cache. This provides a good balance between safety and speed.
If you want the {pve} backup manager to skip a disk when doing a backup of a VM, you can set the No backup option on that disk.
If you want the {pve} storage replication mechanism to skip a disk when starting
a replication job, you can set the Skip replication option on that disk.
As of {pve} 5.0, replication requires the disk images to be on a storage of type
zfspool, so adding a disk image to other storages when the VM has replication
configured requires to skip replication for this disk image.
If your storage supports thin provisioning (see the storage chapter in the {pve} guide), you can activate the Discard option on a drive. With Discard set and a TRIM-enabled guest OS [3], when the VM’s filesystem marks blocks as unused after deleting files, the controller will relay this information to the storage, which will then shrink the disk image accordingly. For the guest to be able to issue TRIM commands, you must enable the Discard option on the drive. Some guest operating systems may also require the SSD Emulation flag to be set. Note that Discard on VirtIO Block drives is only supported on guests using Linux Kernel 5.0 or higher.
If you would like a drive to be presented to the guest as a solid-state drive rather than a rotational hard disk, you can set the SSD emulation option on that drive. There is no requirement that the underlying storage actually be backed by SSDs; this feature can be used with physical media of any type. Note that SSD emulation is not supported on VirtIO Block drives.
The option IO Thread can only be used when using a disk with the VirtIO controller, or with the SCSI controller, when the emulated controller type is VirtIO SCSI single. With IO Thread enabled, QEMU creates one I/O thread per storage controller rather than handling all I/O in the main event loop or vCPU threads. One benefit is better work distribution and utilization of the underlying storage. Another benefit is reduced latency (hangs) in the guest for very I/O-intensive host workloads, since neither the main thread nor a vCPU thread can be blocked by disk I/O.
A CPU socket is a physical slot on a PC motherboard where you can plug a CPU. This CPU can then contain one or many cores, which are independent processing units. Whether you have a single CPU socket with 4 cores, or two CPU sockets with two cores is mostly irrelevant from a performance point of view. However some software licenses depend on the number of sockets a machine has, in that case it makes sense to set the number of sockets to what the license allows you.
Increasing the number of virtual CPUs (cores and sockets) will usually provide a performance improvement though that is heavily dependent on the use of the VM. Multi-threaded applications will of course benefit from a large number of virtual CPUs, as for each virtual cpu you add, QEMU will create a new thread of execution on the host system. If you’re not sure about the workload of your VM, it is usually a safe bet to set the number of Total cores to 2.
|
Note
|
It is perfectly safe if the overall number of cores of all your VMs is greater than the number of cores on the server (for example, 4 VMs each with 4 cores (= total 16) on a machine with only 8 cores). In that case the host system will balance the QEMU execution threads between your server cores, just like if you were running a standard multi-threaded application. However, {pve} will prevent you from starting VMs with more virtual CPU cores than physically available, as this will only bring the performance down due to the cost of context switches. |
cpulimit
In addition to the number of virtual cores, the total available “Host CPU
Time” for the VM can be set with the cpulimit option. It is a floating point
value representing CPU time in percent, so 1.0 is equal to 100%, 2.5 to
250% and so on. If a single process would fully use one single core it would
have 100% CPU Time usage. If a VM with four cores utilizes all its cores
fully it would theoretically use 400%. In reality the usage may be even a bit
higher as QEMU can have additional threads for VM peripherals besides the vCPU
core ones.
This setting can be useful when a VM should have multiple vCPUs because it is running some processes in parallel, but the VM as a whole should not be able to run all vCPUs at 100% at the same time.
For example, suppose you have a virtual machine that would benefit from having 8
virtual CPUs, but you don’t want the VM to be able to max out all 8 cores
running at full load - because that would overload the server and leave other
virtual machines and containers with too little CPU time. To solve this, you
could set cpulimit to 4.0 (=400%). This means that if the VM fully utilizes
all 8 virtual CPUs by running 8 processes simultaneously, each vCPU will receive
a maximum of 50% CPU time from the physical cores. However, if the VM workload
only fully utilizes 4 virtual CPUs, it could still receive up to 100% CPU time
from a physical core, for a total of 400%.
|
Note
|
VMs can, depending on their configuration, use additional threads, such as for networking or IO operations but also live migration. Thus a VM can show up to use more CPU time than just its virtual CPUs could use. To ensure that a VM never uses more CPU time than vCPUs assigned, set the cpulimit to the same value as the total core count. |
cpuunits
With the cpuunits option, nowadays often called CPU shares or CPU weight, you
can control how much CPU time a VM gets compared to other running VMs. It is a
relative weight which defaults to 100 (or 1024 if the host uses legacy
cgroup v1). If you increase this for a VM it will be prioritized by the
scheduler in comparison to other VMs with lower weight.
For example, if VM 100 has set the default 100 and VM 200 was changed to
200, the latter VM 200 would receive twice the CPU bandwidth than the first
VM 100.
For more information see man systemd.resource-control, here CPUQuota
corresponds to cpulimit and CPUWeight to our cpuunits setting. Visit its
Notes section for references and implementation details.
affinity
With the affinity option, you can specify the physical CPU cores that are used to run the VM’s vCPUs. Peripheral VM processes, such as those for I/O, are not affected by this setting. Note that the CPU affinity is not a security feature.
Forcing a CPU affinity can make sense in certain cases but is accompanied by an increase in complexity and maintenance effort. For example, if you want to add more VMs later or migrate VMs to nodes with fewer CPU cores. It can also easily lead to asynchronous and therefore limited system performance if some CPUs are fully utilized while others are almost idle.
The affinity is set through the taskset CLI tool. It accepts the host CPU
numbers (see lscpu) in the List Format from man cpuset. This ASCII decimal
list can contain numbers but also number ranges. For example, the affinity
0-1,8-11 (expanded 0, 1, 8, 9, 10, 11) would allow the VM to run on only
these six specific host cores.
QEMU can emulate a number different of CPU types from 486 to the latest Xeon processors. Each new processor generation adds new features, like hardware assisted 3d rendering, random number generation, memory protection, etc. Also, a current generation can be upgraded through microcode update with bug or security fixes.
Usually you should select for your VM a processor type which closely matches the CPU of the host system, as it means that the host CPU features (also called CPU flags ) will be available in your VMs. If you want an exact match, you can set the CPU type to host in which case the VM will have exactly the same CPU flags as your host system.
This has a downside though. If you want to do a live migration of VMs between different hosts, your VM might end up on a new system with a different CPU type or a different microcode version. If the CPU flags passed to the guest are missing, the QEMU process will stop. To remedy this QEMU has also its own virtual CPU types, that {pve} uses by default.
The backend default is kvm64 which works on essentially all x86_64 host CPUs and the UI default when creating a new VM is x86-64-v2-AES, which requires a host CPU starting from Westmere for Intel or at least a fourth generation Opteron for AMD.
In short:
If you don’t care about live migration or have a homogeneous cluster where all nodes have the same CPU and same microcode version, set the CPU type to host, as in theory this will give your guests maximum performance.
If you care about live migration and security, and you have only Intel CPUs or only AMD CPUs, choose the lowest generation CPU model of your cluster.
If you care about live migration without security, or have mixed Intel/AMD cluster, choose the lowest compatible virtual QEMU CPU type.
|
Note
|
Live migrations between Intel and AMD host CPUs have no guarantee to work. |
QEMU also provide virtual CPU types, compatible with both Intel and AMD host CPUs.
|
Note
|
To mitigate the Spectre vulnerability for virtual CPU types, you need to add the relevant CPU flags, see Meltdown / Spectre related CPU flags. |
Historically, {pve} had the kvm64 CPU model, with CPU flags at the level of Pentium 4 enabled, so performance was not great for certain workloads.
In the summer of 2020, AMD, Intel, Red Hat, and SUSE collaborated to define three x86-64 microarchitecture levels on top of the x86-64 baseline, with modern flags enabled. For details, see the x86-64-ABI specification.
|
Note
|
Some newer distributions like CentOS 9 are now built with x86-64-v2 flags as a minimum requirement. |
-
kvm64 (x86-64-v1): Compatible with Intel CPU >= Pentium 4, AMD CPU >= Phenom.
-
x86-64-v2: Compatible with Intel CPU >= Nehalem, AMD CPU >= Opteron_G3. Added CPU flags compared to x86-64-v1: +cx16, +lahf-lm, +popcnt, +pni, +sse4.1, +sse4.2, +ssse3.
-
x86-64-v2-AES: Compatible with Intel CPU >= Westmere, AMD CPU >= Opteron_G4. Added CPU flags compared to x86-64-v2: +aes.
-
x86-64-v3: Compatible with Intel CPU >= Broadwell, AMD CPU >= EPYC. Added CPU flags compared to x86-64-v2-AES: +avx, +avx2, +bmi1, +bmi2, +f16c, +fma, +movbe, +xsave.
-
x86-64-v4: Compatible with Intel CPU >= Skylake, AMD CPU >= EPYC v4 Genoa. Added CPU flags compared to x86-64-v3: +avx512f, +avx512bw, +avx512cd, +avx512dq, +avx512vl.
You can specify custom CPU types with a configurable set of features. These are
maintained in the configuration file /etc/pve/virtual-guest/cpu-models.conf by
an administrator. See man cpu-models.conf for format details.
Specified custom types can be selected by any user with the Sys.Audit
privilege on /nodes. When configuring a custom CPU type for a VM via the CLI
or API, the name needs to be prefixed with custom-.
There are several CPU flags related to the Meltdown and Spectre vulnerabilities [4] which need to be set manually unless the selected CPU type of your VM already enables them by default.
There are two requirements that need to be fulfilled in order to use these CPU flags:
-
The host CPU(s) must support the feature and propagate it to the guest’s virtual CPU(s)
-
The guest operating system must be updated to a version which mitigates the attacks and is able to utilize the CPU feature
Otherwise you need to set the desired CPU flag of the virtual CPU, either by editing the CPU options in the web UI, or by setting the flags property of the cpu option in the VM configuration file.
For Spectre v1,v2,v4 fixes, your CPU or system vendor also needs to provide a so-called “microcode update” for your CPU, see chapter Firmware Updates. Note that not all affected CPUs can be updated to support spec-ctrl.
To check if the {pve} host is vulnerable, execute the following command as root:
for f in /sys/devices/system/cpu/vulnerabilities/*; do echo "${f##*/} -" $(cat "$f"); done
A community script is also available to detect if the host is still vulnerable. [5]
-
pcid
This reduces the performance impact of the Meltdown (CVE-2017-5754) mitigation called Kernel Page-Table Isolation (KPTI), which effectively hides the Kernel memory from the user space. Without PCID, KPTI is quite an expensive mechanism [6].
To check if the {pve} host supports PCID, execute the following command as root:
# grep ' pcid ' /proc/cpuinfo
If this does not return empty your host’s CPU has support for pcid.
-
spec-ctrl
Required to enable the Spectre v1 (CVE-2017-5753) and Spectre v2 (CVE-2017-5715) fix, in cases where retpolines are not sufficient. Included by default in Intel CPU models with -IBRS suffix. Must be explicitly turned on for Intel CPU models without -IBRS suffix. Requires an updated host CPU microcode (intel-microcode >= 20180425).
-
ssbd
Required to enable the Spectre V4 (CVE-2018-3639) fix. Not included by default in any Intel CPU model. Must be explicitly turned on for all Intel CPU models. Requires an updated host CPU microcode(intel-microcode >= 20180703).
-
ibpb
Required to enable the Spectre v1 (CVE-2017-5753) and Spectre v2 (CVE-2017-5715) fix, in cases where retpolines are not sufficient. Included by default in AMD CPU models with -IBPB suffix. Must be explicitly turned on for AMD CPU models without -IBPB suffix. Requires the host CPU microcode to support this feature before it can be used for guest CPUs.
-
virt-ssbd
Required to enable the Spectre v4 (CVE-2018-3639) fix. Not included by default in any AMD CPU model. Must be explicitly turned on for all AMD CPU models. This should be provided to guests, even if amd-ssbd is also provided, for maximum guest compatibility. Note that this must be explicitly enabled when when using the "host" cpu model, because this is a virtual feature which does not exist in the physical CPUs.
-
amd-ssbd
Required to enable the Spectre v4 (CVE-2018-3639) fix. Not included by default in any AMD CPU model. Must be explicitly turned on for all AMD CPU models. This provides higher performance than virt-ssbd, therefore a host supporting this should always expose this to guests if possible. virt-ssbd should none the less also be exposed for maximum guest compatibility as some kernels only know about virt-ssbd.
-
amd-no-ssb
Recommended to indicate the host is not vulnerable to Spectre V4 (CVE-2018-3639). Not included by default in any AMD CPU model. Future hardware generations of CPU will not be vulnerable to CVE-2018-3639, and thus the guest should be told not to enable its mitigations, by exposing amd-no-ssb. This is mutually exclusive with virt-ssbd and amd-ssbd.
You can also optionally emulate a NUMA [7] architecture in your VMs. The basics of the NUMA architecture mean that instead of having a global memory pool available to all your cores, the memory is spread into local banks close to each socket. This can bring speed improvements as the memory bus is not a bottleneck anymore. If your system has a NUMA architecture [8] we recommend to activate the option, as this will allow proper distribution of the VM resources on the host system. This option is also required to hot-plug cores or RAM in a VM.
If the NUMA option is used, it is recommended to set the number of sockets to the number of nodes of the host system.
Modern operating systems introduced the capability to hot-plug and, to a certain extent, hot-unplug CPUs in a running system. Virtualization allows us to avoid a lot of the (physical) problems real hardware can cause in such scenarios. Still, this is a rather new and complicated feature, so its use should be restricted to cases where its absolutely needed. Most of the functionality can be replicated with other, well tested and less complicated, features, see Resource Limits.
In {pve} the maximal number of plugged CPUs is always cores * sockets.
To start a VM with less than this total core count of CPUs you may use the
vcpus setting, it denotes how many vCPUs should be plugged in at VM start.
Currently only this feature is only supported on Linux, a kernel newer than 3.10 is needed, a kernel newer than 4.7 is recommended.
You can use a udev rule as follow to automatically set new CPUs as online in the guest:
SUBSYSTEM=="cpu", ACTION=="add", TEST=="online", ATTR{online}=="0", ATTR{online}="1"
Save this under /etc/udev/rules.d/ as a file ending in .rules.
Note: CPU hot-remove is machine dependent and requires guest cooperation. The deletion command does not guarantee CPU removal to actually happen, typically it’s a request forwarded to guest OS using target dependent mechanism, such as ACPI on x86/amd64.
For each VM you have the option to set a fixed size memory or asking {pve} to dynamically allocate memory based on the current RAM usage of the host.
When setting memory and minimum memory to the same amount {pve} will simply allocate what you specify to your VM.
Even when using a fixed memory size, the ballooning device gets added to the VM, because it delivers useful information such as how much memory the guest really uses. In general, you should leave ballooning enabled, but if you want to disable it (like for debugging purposes), simply uncheck Ballooning Device or set
balloon: 0
in the configuration.
When setting the minimum memory lower than memory, {pve} will make sure that the minimum amount you specified is always available to the VM, and if RAM usage on the host is below 80%, will dynamically add memory to the guest up to the maximum memory specified.
When the host is running low on RAM, the VM will then release some memory
back to the host, swapping running processes if needed and starting the oom
killer in last resort. The passing around of memory between host and guest is
done via a special balloon kernel driver running inside the guest, which will
grab or release memory pages from the host.
[9]
When multiple VMs use the autoallocate facility, it is possible to set a Shares coefficient which indicates the relative amount of the free host memory that each VM should take. Suppose for instance you have four VMs, three of them running an HTTP server and the last one is a database server. To cache more database blocks in the database server RAM, you would like to prioritize the database VM when spare RAM is available. For this you assign a Shares property of 3000 to the database VM, leaving the other VMs to the Shares default setting of 1000. The host server has 32GB of RAM, and is currently using 16GB, leaving 32 * 80/100 - 16 = 9GB RAM to be allocated to the VMs on top of their configured minimum memory amount. The database VM will benefit from 9 * 3000 / (3000 + 1000 + 1000 + 1000) = 4.5 GB extra RAM and each HTTP server from 1.5 GB.
All Linux distributions released after 2010 have the balloon kernel driver included. For Windows OSes, the balloon driver needs to be added manually and can incur a slowdown of the guest, so we don’t recommend using it on critical systems.
When allocating RAM to your VMs, a good rule of thumb is always to leave 1GB of RAM available to the host.
SEV (Secure Encrypted Virtualization) enables memory encryption per VM using AES-128 encryption and the AMD Secure Processor.
SEV-ES (Secure Encrypted Virtualization-Encrypted State) in addition encrypts all CPU register contents when a VM stops running, to prevent leakage of information to the hypervisor. This feature is very experimental.
Host Requirements:
-
AMD EPYC CPU
-
SEV-ES is only supported on AMD EPYC 7xx2 and newer
-
configure AMD memory encryption in the BIOS settings of the host machine
-
add "kvm_amd.sev=1" to kernel parameters if not enabled by default
-
add "mem_encrypt=on" to kernel parameters if you want to encrypt memory on the host (SME) see https://www.kernel.org/doc/Documentation/x86/amd-memory-encryption.txt
-
maybe increase SWIOTLB see https://github.com/AMDESE/AMDSEV#faq-4
To check if SEV is enabled on the host search for sev in dmesg and print out
the SEV kernel parameter of kvm_amd:
# dmesg | grep -i sev [...] ccp 0000:45:00.1: sev enabled [...] ccp 0000:45:00.1: SEV API: <buildversion> [...] SEV supported: <number> ASIDs [...] SEV-ES supported: <number> ASIDs # cat /sys/module/kvm_amd/parameters/sev Y
Guest Requirements:
-
edk2-OVMF
-
advisable to use Q35
-
The guest operating system must contain SEV-support.
Limitations:
-
Because the memory is encrypted the memory usage on host is always wrong.
-
Operations that involve saving or restoring memory like snapshots & live migration do not work yet or are attackable. https://github.com/PSPReverse/amd-sev-migration-attack
-
PCI passthrough is not supported.
-
SEV-ES is very experimental.
-
QEMU & AMD-SEV documentation is very limited.
Example Configuration:
# qm set <vmid> -amd_sev type=std,no-debug=1,no-key-sharing=1,kernel-hashes=1
The type defines the encryption technology ("type=" is not necessary). Available options are std & es.
The QEMU policy parameter gets calculated with the no-debug and no-key-sharing parameters. These parameters correspond to policy-bit 0 and 1. If type is es the policy-bit 2 is set to 1 so that SEV-ES is enabled. Policy-bit 3 (nosend) is always set to 1 to prevent migration-attacks. For more information on how to calculate the policy see: AMD SEV API Specification Chapter 3
The kernel-hashes option is off per default for backward compatibility with older OVMF images and guests that do not measure the kernel/initrd. See https://lists.gnu.org/archive/html/qemu-devel/2021-11/msg02598.html
Check if SEV is working on the guest
Method 1 - dmesg:
Output should look like this.
# dmesg | grep -i sev AMD Memory Encryption Features active: SEV
Method 2 - MSR 0xc0010131 (MSR_AMD64_SEV):
Output should be 1.
# apt install msr-tools # modprobe msr # rdmsr -a 0xc0010131 1
Links:
Each VM can have many Network interface controllers (NIC), of four different types:
-
Intel E1000 is the default, and emulates an Intel Gigabit network card.
-
the VirtIO paravirtualized NIC should be used if you aim for maximum performance. Like all VirtIO devices, the guest OS should have the proper driver installed.
-
the Realtek 8139 emulates an older 100 MB/s network card, and should only be used when emulating older operating systems ( released before 2002 )
-
the vmxnet3 is another paravirtualized device, which should only be used when importing a VM from another hypervisor.
{pve} will generate for each NIC a random MAC address, so that your VM is addressable on Ethernet networks.
The NIC you added to the VM can follow one of two different models:
-
in the default Bridged mode each virtual NIC is backed on the host by a tap device, ( a software loopback device simulating an Ethernet NIC ). This tap device is added to a bridge, by default vmbr0 in {pve}. In this mode, VMs have direct access to the Ethernet LAN on which the host is located.
-
in the alternative NAT mode, each virtual NIC will only communicate with the QEMU user networking stack, where a built-in router and DHCP server can provide network access. This built-in DHCP will serve addresses in the private 10.0.2.0/24 range. The NAT mode is much slower than the bridged mode, and should only be used for testing. This mode is only available via CLI or the API, but not via the web UI.
You can also skip adding a network device when creating a VM by selecting No network device.
You can overwrite the MTU setting for each VM network device. The option
mtu=1 represents a special case, in which the MTU value will be inherited
from the underlying bridge.
This option is only available for VirtIO network devices.
If you are using the VirtIO driver, you can optionally activate the Multiqueue option. This option allows the guest OS to process networking packets using multiple virtual CPUs, providing an increase in the total number of packets transferred.
When using the VirtIO driver with {pve}, each NIC network queue is passed to the host kernel, where the queue will be processed by a kernel thread spawned by the vhost driver. With this option activated, it is possible to pass multiple network queues to the host kernel for each NIC.
When using Multiqueue, it is recommended to set it to a value equal to the number of vCPUs of your guest. Remember that the number of vCPUs is the number of sockets times the number of cores configured for the VM. You also need to set the number of multi-purpose channels on each VirtIO NIC in the VM with this ethtool command:
ethtool -L ens1 combined X
where X is the number of the number of vCPUs of the VM.
To configure a Windows guest for Multiqueue install the Redhat VirtIO Ethernet Adapter drivers, then adapt the NIC’s configuration as follows. Open the device manager, right click the NIC under "Network adapters", and select "Properties". Then open the "Advanced" tab and select "Receive Side Scaling" from the list on the left. Make sure it is set to "Enabled". Next, navigate to "Maximum number of RSS Queues" in the list and set it to the number of vCPUs of your VM. Once you verified that the settings are correct, click "OK" to confirm them.
You should note that setting the Multiqueue parameter to a value greater than one will increase the CPU load on the host and guest systems as the traffic increases. We recommend to set this option only when the VM has to process a great number of incoming connections, such as when the VM is running as a router, reverse proxy or a busy HTTP server doing long polling.
QEMU can virtualize a few types of VGA hardware. Some examples are:
-
std, the default, emulates a card with Bochs VBE extensions.
-
cirrus, this was once the default, it emulates a very old hardware module with all its problems. This display type should only be used if really necessary [10], for example, if using Windows XP or earlier
-
vmware, is a VMWare SVGA-II compatible adapter.
-
qxl, is the QXL paravirtualized graphics card. Selecting this also enables SPICE (a remote viewer protocol) for the VM.
-
virtio-gl, often named VirGL is a virtual 3D GPU for use inside VMs that can offload workloads to the host GPU without requiring special (expensive) models and drivers and neither binding the host GPU completely, allowing reuse between multiple guests and or the host.
NoteVirGL support needs some extra libraries that aren’t installed by default due to being relatively big and also not available as open source for all GPU models/vendors. For most setups you’ll just need to do: apt install libgl1 libegl1
You can edit the amount of memory given to the virtual GPU, by setting the memory option. This can enable higher resolutions inside the VM, especially with SPICE/QXL.
As the memory is reserved by display device, selecting Multi-Monitor mode
for SPICE (such as qxl2 for dual monitors) has some implications:
-
Windows needs a device for each monitor, so if your ostype is some version of Windows, {pve} gives the VM an extra device per monitor. Each device gets the specified amount of memory.
-
Linux VMs, can always enable more virtual monitors, but selecting a Multi-Monitor mode multiplies the memory given to the device with the number of monitors.
Selecting serialX as display type disables the VGA output, and redirects
the Web Console to the selected serial port. A configured display memory
setting will be ignored in that case.
You can enable the VNC clipboard by setting clipboard to vnc.
# qm set <vmid> -vga <displaytype>,clipboard=vnc
In order to use the clipboard feature, you must first install the
SPICE guest tools. On Debian-based distributions, this can be achieved
by installing spice-vdagent. For other Operating Systems search for it
in the official repositories or see: https://www.spice-space.org/download.html
Once you have installed the spice guest tools, you can use the VNC clipboard
function (e.g. in the noVNC console panel). However, if you’re using
SPICE, virtio or virgl, you’ll need to choose which clipboard to use.
This is because the default SPICE clipboard will be replaced by the
VNC clipboard, if clipboard is set to vnc.
There are two different types of USB passthrough devices:
-
Host USB passthrough
-
SPICE USB passthrough
Host USB passthrough works by giving a VM a USB device of the host. This can either be done via the vendor- and product-id, or via the host bus and port.
The vendor/product-id looks like this: 0123:abcd, where 0123 is the id of the vendor, and abcd is the id of the product, meaning two pieces of the same usb device have the same id.
The bus/port looks like this: 1-2.3.4, where 1 is the bus and 2.3.4 is the port path. This represents the physical ports of your host (depending of the internal order of the usb controllers).
If a device is present in a VM configuration when the VM starts up, but the device is not present in the host, the VM can boot without problems. As soon as the device/port is available in the host, it gets passed through.
|
Warning
|
Using this kind of USB passthrough means that you cannot move a VM online to another host, since the hardware is only available on the host the VM is currently residing. |
The second type of passthrough is SPICE USB passthrough. If you add one or more SPICE USB ports to your VM, you can dynamically pass a local USB device from your SPICE client through to the VM. This can be useful to redirect an input device or hardware dongle temporarily.
It is also possible to map devices on a cluster level, so that they can be properly used with HA and hardware changes are detected and non root users can configure them. See Resource Mapping for details on that.
In order to properly emulate a computer, QEMU needs to use a firmware. Which, on common PCs often known as BIOS or (U)EFI, is executed as one of the first steps when booting a VM. It is responsible for doing basic hardware initialization and for providing an interface to the firmware and hardware for the operating system. By default QEMU uses SeaBIOS for this, which is an open-source, x86 BIOS implementation. SeaBIOS is a good choice for most standard setups.
Some operating systems (such as Windows 11) may require use of an UEFI compatible implementation. In such cases, you must use OVMF instead, which is an open-source UEFI implementation. [11]
There are other scenarios in which the SeaBIOS may not be the ideal firmware to boot from, for example if you want to do VGA passthrough. [12]
If you want to use OVMF, there are several things to consider:
In order to save things like the boot order, there needs to be an EFI Disk. This disk will be included in backups and snapshots, and there can only be one.
You can create such a disk with the following command:
# qm set <vmid> -efidisk0 <storage>:1,format=<format>,efitype=4m,pre-enrolled-keys=1
Where <storage> is the storage where you want to have the disk, and <format> is a format which the storage supports. Alternatively, you can create such a disk through the web interface with Add → EFI Disk in the hardware section of a VM.
The efitype option specifies which version of the OVMF firmware should be used. For new VMs, this should always be 4m, as it supports Secure Boot and has more space allocated to support future development (this is the default in the GUI).
pre-enroll-keys specifies if the efidisk should come pre-loaded with distribution-specific and Microsoft Standard Secure Boot keys. It also enables Secure Boot by default (though it can still be disabled in the OVMF menu within the VM).
|
Note
|
If you want to start using Secure Boot in an existing VM (that still uses
a 2m efidisk), you need to recreate the efidisk. To do so, delete the old one
(qm set <vmid> -delete efidisk0) and add a new one as described above. This
will reset any custom configurations you have made in the OVMF menu!
|
When using OVMF with a virtual display (without VGA passthrough), you need to set the client resolution in the OVMF menu (which you can reach with a press of the ESC button during boot), or you have to choose SPICE as the display type.
A Trusted Platform Module is a device which stores secret data - such as encryption keys - securely and provides tamper-resistance functions for validating system boot.
Certain operating systems (such as Windows 11) require such a device to be attached to a machine (be it physical or virtual).
A TPM is added by specifying a tpmstate volume. This works similar to an efidisk, in that it cannot be changed (only removed) once created. You can add one via the following command:
# qm set <vmid> -tpmstate0 <storage>:1,version=<version>
Where <storage> is the storage you want to put the state on, and <version> is either v1.2 or v2.0. You can also add one via the web interface, by choosing Add → TPM State in the hardware section of a VM.
The v2.0 TPM spec is newer and better supported, so unless you have a specific implementation that requires a v1.2 TPM, it should be preferred.
|
Note
|
Compared to a physical TPM, an emulated one does not provide any real security benefits. The point of a TPM is that the data on it cannot be modified easily, except via commands specified as part of the TPM spec. Since with an emulated device the data storage happens on a regular volume, it can potentially be edited by anyone with access to it. |
You can add an Inter-VM shared memory device (ivshmem), which allows one to
share memory between the host and a guest, or also between multiple guests.
To add such a device, you can use qm:
# qm set <vmid> -ivshmem size=32,name=foo
Where the size is in MiB. The file will be located under
/dev/shm/pve-shm-$name (the default name is the vmid).
|
Note
|
Currently the device will get deleted as soon as any VM using it got shutdown or stopped. Open connections will still persist, but new connections to the exact same device cannot be made anymore. |
A use case for such a device is the Looking Glass [13] project, which enables high performance, low-latency display mirroring between host and guest.
To add an audio device run the following command:
qm set <vmid> -audio0 device=<device>
Supported audio devices are:
-
ich9-intel-hda: Intel HD Audio Controller, emulates ICH9 -
intel-hda: Intel HD Audio Controller, emulates ICH6 -
AC97: Audio Codec '97, useful for older operating systems like Windows XP
There are two backends available:
-
spice
-
none
The spice backend can be used in combination with SPICE while the none backend can be useful if an audio device is needed in the VM for some software to work. To use the physical audio device of the host use device passthrough (see PCI Passthrough and USB Passthrough). Remote protocols like Microsoft’s RDP have options to play sound.
A RNG (Random Number Generator) is a device providing entropy (randomness) to a system. A virtual hardware-RNG can be used to provide such entropy from the host system to a guest VM. This helps to avoid entropy starvation problems in the guest (a situation where not enough entropy is available and the system may slow down or run into problems), especially during the guests boot process.
To add a VirtIO-based emulated RNG, run the following command:
qm set <vmid> -rng0 source=<source>[,max_bytes=X,period=Y]
source specifies where entropy is read from on the host and has to be one of
the following:
-
/dev/urandom: Non-blocking kernel entropy pool (preferred) -
/dev/random: Blocking kernel pool (not recommended, can lead to entropy starvation on the host system) -
/dev/hwrng: To pass through a hardware RNG attached to the host (if multiple are available, the one selected in/sys/devices/virtual/misc/hw_random/rng_currentwill be used)
A limit can be specified via the max_bytes and period parameters, they are
read as max_bytes per period in milliseconds. However, it does not represent
a linear relationship: 1024B/1000ms would mean that up to 1 KiB of data becomes
available on a 1 second timer, not that 1 KiB is streamed to the guest over the
course of one second. Reducing the period can thus be used to inject entropy
into the guest at a faster rate.
By default, the limit is set to 1024 bytes per 1000 ms (1 KiB/s). It is
recommended to always use a limiter to avoid guests using too many host
resources. If desired, a value of 0 for max_bytes can be used to disable
all limits.
QEMU can tell the guest which devices it should boot from, and in which order.
This can be specified in the config via the boot property, for example:
boot: order=scsi0;net0;hostpci0
This way, the guest would first attempt to boot from the disk scsi0, if that
fails, it would go on to attempt network boot from net0, and in case that
fails too, finally attempt to boot from a passed through PCIe device (seen as
disk in case of NVMe, otherwise tries to launch into an option ROM).
On the GUI you can use a drag-and-drop editor to specify the boot order, and use the checkbox to enable or disable certain devices for booting altogether.
|
Note
|
If your guest uses multiple disks to boot the OS or load the bootloader, all of them must be marked as bootable (that is, they must have the checkbox enabled or appear in the list in the config) for the guest to be able to boot. This is because recent SeaBIOS and OVMF versions only initialize disks if they are marked bootable. |
In any case, even devices not appearing in the list or having the checkmark disabled will still be available to the guest, once it’s operating system has booted and initialized them. The bootable flag only affects the guest BIOS and bootloader.
After creating your VMs, you probably want them to start automatically when the host system boots. For this you need to select the option Start at boot from the Options Tab of your VM in the web interface, or set it with the following command:
# qm set <vmid> -onboot 1
In some case you want to be able to fine tune the boot order of your VMs, for instance if one of your VM is providing firewalling or DHCP to other guest systems. For this you can use the following parameters:
-
Start/Shutdown order: Defines the start order priority. For example, set it to 1 if you want the VM to be the first to be started. (We use the reverse startup order for shutdown, so a machine with a start order of 1 would be the last to be shut down). If multiple VMs have the same order defined on a host, they will additionally be ordered by VMID in ascending order.
-
Startup delay: Defines the interval between this VM start and subsequent VMs starts. For example, set it to 240 if you want to wait 240 seconds before starting other VMs.
-
Shutdown timeout: Defines the duration in seconds {pve} should wait for the VM to be offline after issuing a shutdown command. By default this value is set to 180, which means that {pve} will issue a shutdown request and wait 180 seconds for the machine to be offline. If the machine is still online after the timeout it will be stopped forcefully.
|
Note
|
VMs managed by the HA stack do not follow the start on boot and boot order options currently. Those VMs will be skipped by the startup and shutdown algorithm as the HA manager itself ensures that VMs get started and stopped. |
Please note that machines without a Start/Shutdown order parameter will always start after those where the parameter is set. Further, this parameter can only be enforced between virtual machines running on the same host, not cluster-wide.
If you require a delay between the host boot and the booting of the first VM, see the section on Proxmox VE Node Management.
The QEMU Guest Agent is a service which runs inside the VM, providing a communication channel between the host and the guest. It is used to exchange information and allows the host to issue commands to the guest.
For example, the IP addresses in the VM summary panel are fetched via the guest agent.
Or when starting a backup, the guest is told via the guest agent to sync outstanding writes via the fs-freeze and fs-thaw commands.
For the guest agent to work properly the following steps must be taken:
-
install the agent in the guest and make sure it is running
-
enable the communication via the agent in {pve}
For most Linux distributions, the guest agent is available. The package is
usually named qemu-guest-agent.
For Windows, it can be installed from the Fedora VirtIO driver ISO.
Communication from {pve} with the guest agent can be enabled in the VM’s Options panel. A fresh start of the VM is necessary for the changes to take effect.
It is possible to enable the Run guest-trim option. With this enabled, {pve} will issue a trim command to the guest after the following operations that have the potential to write out zeros to the storage:
-
moving a disk to another storage
-
live migrating a VM to another node with local storage
On a thin provisioned storage, this can help to free up unused space.
|
Note
|
There is a caveat with ext4 on Linux, because it uses an in-memory optimization to avoid issuing duplicate TRIM requests. Since the guest doesn’t know about the change in the underlying storage, only the first guest-trim will run as expected. Subsequent ones, until the next reboot, will only consider parts of the filesystem that changed since then. |
By default, guest filesystems are synced via the fs-freeze QEMU Guest Agent Command when a backup is performed, to provide consistency.
On Windows guests, some applications might handle consistent backups themselves by hooking into the Windows VSS (Volume Shadow Copy Service) layer, a fs-freeze then might interfere with that. For example, it has been observed that calling fs-freeze with some SQL Servers triggers VSS to call the SQL Writer VSS module in a mode that breaks the SQL Server backup chain for differential backups.
For such setups you can configure {pve} to not issue a freeze-and-thaw cycle on
backup by setting the freeze-fs-on-backup QGA option to 0. This can also be
done via the GUI with the Freeze/thaw guest filesystems on backup for
consistency option.
|
Important
|
Disabling this option can potentially lead to backups with inconsistent filesystems and should therefore only be disabled if you know what you are doing. |
Make sure the guest agent is installed and running.
Once the guest agent is enabled, {pve} will send power commands like shutdown via the guest agent. If the guest agent is not running, commands cannot get executed properly and the shutdown command will run into a timeout.
SPICE Enhancements are optional features that can improve the remote viewer experience.
To enable them via the GUI go to the Options panel of the virtual machine. Run the following command to enable them via the CLI:
qm set <vmid> -spice_enhancements foldersharing=1,videostreaming=all
|
Note
|
To use these features the Display of the virtual machine must be set to SPICE (qxl). |
Share a local folder with the guest. The spice-webdavd daemon needs to be
installed in the guest. It makes the shared folder available through a local
WebDAV server located at http://localhost:9843.
For Windows guests the installer for the Spice WebDAV daemon can be downloaded from the official SPICE website.
Most Linux distributions have a package called spice-webdavd that can be
installed.
To share a folder in Virt-Viewer (Remote Viewer) go to File → Preferences. Select the folder to share and then enable the checkbox.
|
Note
|
Folder sharing currently only works in the Linux version of Virt-Viewer. |
|
Caution
|
Experimental! Currently this feature does not work reliably. |
Fast refreshing areas are encoded into a video stream. Two options exist:
-
all: Any fast refreshing area will be encoded into a video stream.
-
filter: Additional filters are used to decide if video streaming should be used (currently only small window surfaces are skipped).
A general recommendation if video streaming should be enabled and which option to choose from cannot be given. Your mileage may vary depending on the specific circumstances.
Make sure the WebDAV service is enabled and running in the guest. On Windows it is called Spice webdav proxy. In Linux the name is spice-webdavd but can be different depending on the distribution.
If the service is running, check the WebDAV server by opening http://localhost:9843 in a browser in the guest.
It can help to restart the SPICE session.
If you have a cluster, you can migrate your VM to another host with
# qm migrate <vmid> <target>
There are generally two mechanisms for this
-
Online Migration (aka Live Migration)
-
Offline Migration
If your VM is running and no locally bound resources are configured (such as
devices that are passed through), you can initiate a live migration with the --online
flag in the qm migration command evocation. The web interface defaults to
live migration when the VM is running.
Online migration first starts a new QEMU process on the target host with the incoming flag, which performs only basic initialization with the guest vCPUs still paused and then waits for the guest memory and device state data streams of the source Virtual Machine. All other resources, such as disks, are either shared or got already sent before runtime state migration of the VMs begins; so only the memory content and device state remain to be transferred.
Once this connection is established, the source begins asynchronously sending the memory content to the target. If the guest memory on the source changes, those sections are marked dirty and another pass is made to send the guest memory data. This loop is repeated until the data difference between running source VM and incoming target VM is small enough to be sent in a few milliseconds, because then the source VM can be paused completely, without a user or program noticing the pause, so that the remaining data can be sent to the target, and then unpause the targets VM’s CPU to make it the new running VM in well under a second.
For Live Migration to work, there are some things required:
-
The VM has no local resources that cannot be migrated. For example, PCI or USB devices that are passed through currently block live-migration. Local Disks, on the other hand, can be migrated by sending them to the target just fine.
-
The hosts are located in the same {pve} cluster.
-
The hosts have a working (and reliable) network connection between them.
-
The target host must have the same, or higher versions of the {pve} packages. Although it can sometimes work the other way around, this cannot be guaranteed.
-
The hosts have CPUs from the same vendor with similar capabilities. Different vendor might work depending on the actual models and VMs CPU type configured, but it cannot be guaranteed - so please test before deploying such a setup in production.
If you have local resources, you can still migrate your VMs offline as long as all disk are on storage defined on both hosts. Migration then copies the disks to the target host over the network, as with online migration. Note that any hardware passthrough configuration may need to be adapted to the device location on the target host.
VM installation is usually done using an installation media (CD-ROM) from the operating system vendor. Depending on the OS, this can be a time consuming task one might want to avoid.
An easy way to deploy many VMs of the same type is to copy an existing VM. We use the term clone for such copies, and distinguish between linked and full clones.
- Full Clone
-
The result of such copy is an independent VM. The new VM does not share any storage resources with the original.
It is possible to select a Target Storage, so one can use this to migrate a VM to a totally different storage. You can also change the disk image Format if the storage driver supports several formats.
NoteA full clone needs to read and copy all VM image data. This is usually much slower than creating a linked clone. Some storage types allows to copy a specific Snapshot, which defaults to the current VM data. This also means that the final copy never includes any additional snapshots from the original VM.
- Linked Clone
-
Modern storage drivers support a way to generate fast linked clones. Such a clone is a writable copy whose initial contents are the same as the original data. Creating a linked clone is nearly instantaneous, and initially consumes no additional space.
They are called linked because the new image still refers to the original. Unmodified data blocks are read from the original image, but modification are written (and afterwards read) from a new location. This technique is called Copy-on-write.
This requires that the original volume is read-only. With {pve} one can convert any VM into a read-only Template). Such templates can later be used to create linked clones efficiently.
NoteYou cannot delete an original template while linked clones exist. It is not possible to change the Target storage for linked clones, because this is a storage internal feature.
The Target node option allows you to create the new VM on a different node. The only restriction is that the VM is on shared storage, and that storage is also available on the target node.
To avoid resource conflicts, all network interface MAC addresses get randomized, and we generate a new UUID for the VM BIOS (smbios1) setting.
One can convert a VM into a Template. Such templates are read-only, and you can use them to create linked clones.
|
Note
|
It is not possible to start templates, because this would modify the disk images. If you want to change the template, create a linked clone and modify that. |
{pve} supports Virtual Machine Generation ID (vmgenid) [14] for virtual machines. This can be used by the guest operating system to detect any event resulting in a time shift event, for example, restoring a backup or a snapshot rollback.
When creating new VMs, a vmgenid will be automatically generated and saved in its configuration file.
To create and add a vmgenid to an already existing VM one can pass the special value ‘1’ to let {pve} autogenerate one or manually set the UUID [15] by using it as value, for example:
# qm set VMID -vmgenid 1 # qm set VMID -vmgenid 00000000-0000-0000-0000-000000000000
|
Note
|
The initial addition of a vmgenid device to an existing VM, may result in the same effects as a change on snapshot rollback, backup restore, etc., has as the VM can interpret this as generation change. |
In the rare case the vmgenid mechanism is not wanted one can pass ‘0’ for its value on VM creation, or retroactively delete the property in the configuration with:
# qm set VMID -delete vmgenid
The most prominent use case for vmgenid are newer Microsoft Windows operating systems, which use it to avoid problems in time sensitive or replicate services (such as databases or domain controller [16]) on snapshot rollback, backup restore or a whole VM clone operation.
Importing existing virtual machines from foreign hypervisors or other {pve} clusters can be achieved through various methods, the most common ones are:
-
Using the native import wizard, which utilizes the import content type, such as provided by the ESXi special storage.
-
Performing a backup on the source and then restoring on the target. This method works best when migrating from another {pve} instance.
-
using the OVF-specific import command of the
qmcommand-line tool.
If you import VMs to {pve} from other hypervisors, it’s recommended to familiarize yourself with the concepts of {pve}.
{pve} provides an integrated VM importer using the storage plugin system for native integration into the API and web-based user interface. You can use this to import the VM as a whole, with most of its config mapped to {pve}'s config model and reduced downtime.
|
Note
|
The import wizard was added during the {pve} 8.2 development cycle and is in tech preview state. While it’s already promising and working stable, it’s still under active development. |
To use the import wizard you have to first set up a new storage for an import source, you can do so on the web-interface under Datacenter → Storage → Add.
Then you can select the new storage in the resource tree and use the Virtual Guests content tab to see all available guests that can be imported.
Select one and use the Import button (or double-click) to open the import wizard. You can modify a subset of the available options here and then start the import. Please note that you can do more advanced modifications after the import finished.
|
Tip
|
The ESXi import wizard has been tested with ESXi versions 6.5 through 8.0. Note that guests using vSAN storage cannot be directly imported directly; their disks must first be moved to another storage. While it is possible to use a vCenter as the import source, performance is dramatically degraded (5 to 10 times slower). |
For a step-by-step guide and tips for how to adapt the virtual guest to the new hyper-visor see our migrate to {pve} wiki article.
To import OVA/OVF files, you first need a File-based storage with the import content type. On this storage, there will be an import folder where you can put OVA files or OVF files with the corresponding images in a flat structure. Alternatively you can use the web UI to upload or download OVA files directly. You can then use the web UI to select those and use the import wizard to import the guests.
For OVA files, there is additional space needed to temporarily extract the image. This needs a file-based storage that has the images content type configured. By default the source storage is selected for this, but you can specify a Import Working Storage on which the images will be extracted before importing to the actual target storage.
|
Note
|
Since OVA/OVF file structure and content are not always well maintained or defined, it might be necessary to adapt some guest settings manually. For example the SCSI controller type is almost never defined in OVA/OVF files, but the default is unbootable with OVMF (UEFI), so you should select Virtio SCSI or VMware PVSCSI for these cases. |
A VM export from a foreign hypervisor takes usually the form of one or more disk
images, with a configuration file describing the settings of the VM (RAM,
number of cores).
The disk images can be in the vmdk format, if the disks come from
VMware or VirtualBox, or qcow2 if the disks come from a KVM hypervisor.
The most popular configuration format for VM exports is the OVF standard, but in
practice interoperation is limited because many settings are not implemented in
the standard itself, and hypervisors export the supplementary information
in non-standard extensions.
Besides the problem of format, importing disk images from other hypervisors may fail if the emulated hardware changes too much from one hypervisor to another. Windows VMs are particularly concerned by this, as the OS is very picky about any changes of hardware. This problem may be solved by installing the MergeIDE.zip utility available from the Internet before exporting and choosing a hard disk type of IDE before booting the imported Windows VM.
Finally there is the question of paravirtualized drivers, which improve the speed of the emulated system and are specific to the hypervisor. GNU/Linux and other free Unix OSes have all the necessary drivers installed by default and you can switch to the paravirtualized drivers right after importing the VM. For Windows VMs, you need to install the Windows paravirtualized drivers by yourself.
GNU/Linux and other free Unix can usually be imported without hassle. Note that we cannot guarantee a successful import/export of Windows VMs in all cases due to the problems above.
Microsoft provides Virtual Machines downloads to get started with Windows development.We are going to use one of these to demonstrate the OVF import feature.
After getting informed about the user agreement, choose the Windows 10 Enterprise (Evaluation - Build) for the VMware platform, and download the zip.
Using the unzip utility or any archiver of your choice, unpack the zip,
and copy via ssh/scp the ovf and vmdk files to your {pve} host.
This will create a new virtual machine, using cores, memory and
VM name as read from the OVF manifest, and import the disks to the local-lvm
storage. You have to configure the network manually.
# qm importovf 999 WinDev1709Eval.ovf local-lvm
The VM is ready to be started.
You can also add an existing disk image to a VM, either coming from a foreign hypervisor, or one that you created yourself.
Suppose you created a Debian/Ubuntu disk image with the vmdebootstrap tool:
vmdebootstrap --verbose \ --size 10GiB --serial-console \ --grub --no-extlinux \ --package openssh-server \ --package avahi-daemon \ --package qemu-guest-agent \ --hostname vm600 --enable-dhcp \ --customize=./copy_pub_ssh.sh \ --sparse --image vm600.raw
You can now create a new target VM, importing the image to the storage pvedir
and attaching it to the VM’s SCSI controller:
# qm create 600 --net0 virtio,bridge=vmbr0 --name vm600 --serial0 socket \ --boot order=scsi0 --scsihw virtio-scsi-pci --ostype l26 \ --scsi0 pvedir:0,import-from=/path/to/dir/vm600.raw
The VM is ready to be started.
You can add a hook script to VMs with the config property hookscript.
# qm set 100 --hookscript local:snippets/hookscript.pl
It will be called during various phases of the guests lifetime.
For an example and documentation see the example script under
/usr/share/pve-docs/examples/guest-example-hookscript.pl.
You can suspend a VM to disk with the GUI option Hibernate or with
# qm suspend ID --todisk
That means that the current content of the memory will be saved onto disk and the VM gets stopped. On the next start, the memory content will be loaded and the VM can continue where it was left off.
If no target storage for the memory is given, it will be automatically chosen, the first of:
-
The storage
vmstatestoragefrom the VM config. -
The first shared storage from any VM disk.
-
The first non-shared storage from any VM disk.
-
The storage
localas a fallback.
When using or referencing local resources (e.g. address of a pci device), using the raw address or id is sometimes problematic, for example:
-
when using HA, a different device with the same id or path may exist on the target node, and if one is not careful when assigning such guests to HA groups, the wrong device could be used, breaking configurations.
-
changing hardware can change ids and paths, so one would have to check all assigned devices and see if the path or id is still correct.
To handle this better, one can define cluster wide resource mappings, such that a resource has a cluster unique, user selected identifier which can correspond to different devices on different hosts. With this, HA won’t start a guest with a wrong device, and hardware changes can be detected.
Creating such a mapping can be done with the {pve} web GUI under Datacenter
in the relevant tab in the Resource Mappings category, or on the cli with
# pvesh create /cluster/mapping/<type> <options>
Where <type> is the hardware type (currently either pci or usb) and
<options> are the device mappings and other configuration parameters.
Note that the options must include a map property with all identifying properties of that hardware, so that it’s possible to verify the hardware did not change and the correct device is passed through.
For example to add a PCI device as device1 with the path 0000:01:00.0 that
has the device id 0001 and the vendor id 0002 on the node node1, and
0000:02:00.0 on node2 you can add it with:
# pvesh create /cluster/mapping/pci --id device1 \ --map node=node1,path=0000:01:00.0,id=0002:0001 \ --map node=node2,path=0000:02:00.0,id=0002:0001
You must repeat the map parameter for each node where that device should have
a mapping (note that you can currently only map one USB device per node per
mapping).
Using the GUI makes this much easier, as the correct properties are automatically picked up and sent to the API.
It’s also possible for PCI devices to provide multiple devices per node with multiple map properties for the nodes. If such a device is assigned to a guest, the first free one will be used when the guest is started. The order of the paths given is also the order in which they are tried, so arbitrary allocation policies can be implemented.
This is useful for devices with SR-IOV, since some times it is not important which exact virtual function is passed through.
You can assign such a device to a guest either with the GUI or with
# qm set ID -hostpci0 <name>
for PCI devices, or
# qm set <vmid> -usb0 <name>
for USB devices.
Where <vmid> is the guests id and <name> is the chosen name for the created
mapping. All usual options for passing through the devices are allowed, such as
mdev.
To create mappings Mapping.Modify on /mapping/<type>/<name> is necessary
(where <type> is the device type and <name> is the name of the mapping).
To use these mappings, Mapping.Use on /mapping/<type>/<name> is necessary
(in addition to the normal guest privileges to edit the configuration).
qm is the tool to manage QEMU/KVM virtual machines on {pve}. You can create and destroy virtual machines, and control execution (start/stop/suspend/resume). Besides that, you can use qm to set parameters in the associated config file. It is also possible to create and delete virtual disks.
Using an iso file uploaded on the local storage, create a VM with a 4 GB IDE disk on the local-lvm storage
# qm create 300 -ide0 local-lvm:4 -net0 e1000 -cdrom local:iso/proxmox-mailgateway_2.1.iso
Start the new VM
# qm start 300
Send a shutdown request, then wait until the VM is stopped.
# qm shutdown 300 && qm wait 300
Same as above, but only wait for 40 seconds.
# qm shutdown 300 && qm wait 300 -timeout 40
If the VM does not shut down, force-stop it and overrule any running shutdown tasks. As stopping VMs may incur data loss, use it with caution.
# qm stop 300 -overrule-shutdown 1
Destroying a VM always removes it from Access Control Lists and it always removes the firewall configuration of the VM. You have to activate --purge, if you want to additionally remove the VM from replication jobs, backup jobs and HA resource configurations.
# qm destroy 300 --purge
Move a disk image to a different storage.
# qm move-disk 300 scsi0 other-storage
Reassign a disk image to a different VM. This will remove the disk scsi1 from
the source VM and attaches it as scsi3 to the target VM. In the background
the disk image is being renamed so that the name matches the new owner.
# qm move-disk 300 scsi1 --target-vmid 400 --target-disk scsi3
VM configuration files are stored inside the Proxmox cluster file
system, and can be accessed at /etc/pve/qemu-server/<VMID>.conf.
Like other files stored inside /etc/pve/, they get automatically
replicated to all other cluster nodes.
|
Note
|
VMIDs < 100 are reserved for internal purposes, and VMIDs need to be unique cluster wide. |
boot: order=virtio0;net0 cores: 1 sockets: 1 memory: 512 name: webmail ostype: l26 net0: e1000=EE:D2:28:5F:B6:3E,bridge=vmbr0 virtio0: local:vm-100-disk-1,size=32G
Those configuration files are simple text files, and you can edit them
using a normal text editor (vi, nano, …). This is sometimes
useful to do small corrections, but keep in mind that you need to
restart the VM to apply such changes.
For that reason, it is usually better to use the qm command to
generate and modify those files, or do the whole thing using the GUI.
Our toolkit is smart enough to instantaneously apply most changes to
running VM. This feature is called "hot plug", and there is no
need to restart the VM in that case.
VM configuration files use a simple colon separated key/value format. Each line has the following format:
# this is a comment OPTION: value
Blank lines in those files are ignored, and lines starting with a #
character are treated as comments and are also ignored.
When you create a snapshot, qm stores the configuration at snapshot
time into a separate snapshot section within the same configuration
file. For example, after creating a snapshot called “testsnapshot”,
your configuration file will look like this:
memory: 512 swap: 512 parent: testsnaphot ... [testsnaphot] memory: 512 swap: 512 snaptime: 1457170803 ...
There are a few snapshot related properties like parent and
snaptime. The parent property is used to store the parent/child
relationship between snapshots. snaptime is the snapshot creation
time stamp (Unix epoch).
You can optionally save the memory of a running VM with the option vmstate.
For details about how the target storage gets chosen for the VM state, see
State storage selection in the chapter
Hibernation.
Online migrations, snapshots and backups (vzdump) set a lock to prevent
incompatible concurrent actions on the affected VMs. Sometimes you need to
remove such a lock manually (for example after a power failure).
# qm unlock <vmid>
|
Caution
|
Only do that if you are sure the action which set the lock is no longer running. |
numactl --hardware | grep available returns more than one node, then your host system has a NUMA architecture