Difference between revisions of "QEMU"

From ArchWiki
Jump to navigation Jump to search
(→‎Graphic card: +text mode graphics card emulation for using it in any text termnal)
(→‎User-mode networking: ping warning: how to make ping work, point to wiki article)
Line 523: Line 523:
 
By default, without any {{ic|-netdev}} arguments, QEMU will use user-mode networking with a built-in DHCP server. Your virtual machines will be assigned an IP address when they run their DHCP client, and they will be able to access the physical host's network through IP masquerading done by QEMU.
 
By default, without any {{ic|-netdev}} arguments, QEMU will use user-mode networking with a built-in DHCP server. Your virtual machines will be assigned an IP address when they run their DHCP client, and they will be able to access the physical host's network through IP masquerading done by QEMU.
  
{{warning|This only works with the TCP and UDP protocols, so ICMP, including {{ic|ping}}, will not work. Do not use {{ic|ping}} to test network connectivity.}}
+
{{warning|This only works with the TCP and UDP protocols, so ICMP, including {{ic|ping}}, will not work. Do not use {{ic|ping}} to test network connectivity. To make ping work in the guest see [[Sysctl#Allow_unprivileged_users_to_create_IPPROTO_ICMP_sockets]].}}
  
 
This default configuration allows your virtual machines to easily access the Internet, provided that the host is connected to it, but the virtual machines will not be directly visible on the external network, nor will virtual machines be able to talk to each other if you start up more than one concurrently.
 
This default configuration allows your virtual machines to easily access the Internet, provided that the host is connected to it, but the virtual machines will not be directly visible on the external network, nor will virtual machines be able to talk to each other if you start up more than one concurrently.

Revision as of 14:11, 9 August 2019

According to the QEMU about page, "QEMU is a generic and open source machine emulator and virtualizer."

When used as a machine emulator, QEMU can run OSes and programs made for one machine (e.g. an ARM board) on a different machine (e.g. your x86 PC). By using dynamic translation, it achieves very good performance.

QEMU can use other hypervisors like Xen or KVM to use CPU extensions (HVM) for virtualization. When used as a virtualizer, QEMU achieves near native performances by executing the guest code directly on the host CPU.

Contents

Installation

Install the qemu package (or qemu-headless for the version without GUI) and below optional packages for your needs:

Graphical front-ends for QEMU

Unlike other virtualization programs such as VirtualBox and VMware, QEMU does not provide a GUI to manage virtual machines (other than the window that appears when running a virtual machine), nor does it provide a way to create persistent virtual machines with saved settings. All parameters to run a virtual machine must be specified on the command line at every launch, unless you have created a custom script to start your virtual machine(s).

Libvirt provides a convenient way to manage QEMU virtual machines. See list of libvirt clients for available front-ends.

Other GUI front-ends for QEMU:

  • AQEMU — QEMU GUI written in Qt5.
https://github.com/tobimensch/aqemu || aqemuAUR
  • QtEmu — Graphical user interface for QEMU written in Qt4.
https://qtemu.org/ || qtemuAUR

Creating new virtualized system

Creating a hard disk image

Tango-inaccurate.pngThe factual accuracy of this article or section is disputed.Tango-inaccurate.png

Reason: If I get the man page right the raw format only allocates the full size if the filesystem does not support "holes" or it is explicitly told to preallocate. See man qemu-img in section Notes. (Discuss in Talk:QEMU#)
Tip: See the QEMU Wikibook for more information on QEMU images.

To run QEMU you will need a hard disk image, unless you are booting a live system from CD-ROM or the network (and not doing so to install an operating system to a hard disk image). A hard disk image is a file which stores the contents of the emulated hard disk.

A hard disk image can be raw, so that it is literally byte-by-byte the same as what the guest sees, and will always use the full capacity of the guest hard drive on the host. This method provides the least I/O overhead, but can waste a lot of space, as not-used space on the guest cannot be used on the host.

Alternatively, the hard disk image can be in a format such as qcow2 which only allocates space to the image file when the guest operating system actually writes to those sectors on its virtual hard disk. The image appears as the full size to the guest operating system, even though it may take up only a very small amount of space on the host system. This image format also supports QEMU snapshotting functionality (see #Creating and managing snapshots via the monitor console for details). However, using this format instead of raw will likely affect performance.

QEMU provides the qemu-img command to create hard disk images. For example to create a 4 GB image in the raw format:

$ qemu-img create -f raw image_file 4G

You may use -f qcow2 to create a qcow2 disk instead.

Note: You can also simply create a raw image by creating a file of the needed size using dd or fallocate.
Warning: If you store the hard disk images on a Btrfs file system, you should consider disabling Copy-on-Write for the directory before creating any images.

Overlay storage images

You can create a storage image once (the 'backing' image) and have QEMU keep mutations to this image in an overlay image. This allows you to revert to a previous state of this storage image. You could revert by creating a new overlay image at the time you wish to revert, based on the original backing image.

To create an overlay image, issue a command like:

$ qemu-img create -o backing_file=img1.raw,backing_fmt=raw -f qcow2 img1.cow

After that you can run your QEMU VM as usual (see #Running virtualized system):

$ qemu-system-x86_64 img1.cow

The backing image will then be left intact and mutations to this storage will be recorded in the overlay image file.

When the path to the backing image changes, repair is required.

Warning: The backing image's absolute filesystem path is stored in the (binary) overlay image file. Changing the backing image's path requires some effort.

Make sure that the original backing image's path still leads to this image. If necessary, make a symbolic link at the original path to the new path. Then issue a command like:

$ qemu-img rebase -b /new/img1.raw /new/img1.cow

At your discretion, you may alternatively perform an 'unsafe' rebase where the old path to the backing image is not checked:

$ qemu-img rebase -u -b /new/img1.raw /new/img1.cow

Resizing an image

Warning: Resizing an image containing an NTFS boot file system could make the operating system installed on it unbootable. It is recommended to create a backup first.

The qemu-img executable has the resize option, which enables easy resizing of a hard drive image. It works for both raw and qcow2. For example, to increase image space by 10 GB, run:

$ qemu-img resize disk_image +10G

After enlarging the disk image, you must use file system and partitioning tools inside the virtual machine to actually begin using the new space. When shrinking a disk image, you must first reduce the allocated file systems and partition sizes using the file system and partitioning tools inside the virtual machine and then shrink the disk image accordingly, otherwise shrinking the disk image will result in data loss! For a Windows guest, open the "create and format hard disk partitions" control panel.

Converting an image

You can convert an image to other formats using qemu-img convert. This example shows how to convert a raw image to qcow2:

$ qemu-img convert -f raw -O qcow2 input.img output.qcow2

This will not remove the original input file.

Preparing the installation media

To install an operating system into your disk image, you need the installation medium (e.g. optical disk, USB-drive, or ISO image) for the operating system. The installation medium should not be mounted because QEMU accesses the media directly.

Tip: If using an optical disk, it is a good idea to first dump the media to a file because this both improves performance and does not require you to have direct access to the devices (that is, you can run QEMU as a regular user without having to change access permissions on the media's device file). For example, if the CD-ROM device node is named /dev/cdrom, you can dump it to a file with the command:
$ dd if=/dev/cdrom of=cd_image.iso bs=4k

Installing the operating system

This is the first time you will need to start the emulator. To install the operating system on the disk image, you must attach both the disk image and the installation media to the virtual machine, and have it boot from the installation media.

For example on i386 guests, to install from a bootable ISO file as CD-ROM and a raw disk image:

$ qemu-system-x86_64 -cdrom iso_image -boot order=d -drive file=disk_image,format=raw

See qemu(1) for more information about loading other media types (such as floppy, disk images or physical drives) and #Running virtualized system for other useful options.

After the operating system has finished installing, the QEMU image can be booted directly (see #Running virtualized system).

Note: By default only 128 MB of memory is assigned to the machine. The amount of memory can be adjusted with the -m switch, for example -m 512M or -m 2G.
Tip:
  • Instead of specifying -boot order=x, some users may feel more comfortable using a boot menu: -boot menu=on, at least during configuration and experimentation.
  • If you need to replace floppies or CDs as part of the installation process, you can use the QEMU machine monitor (press Ctrl+Alt+2 in the virtual machine's window) to remove and attach storage devices to a virtual machine. Type info block to see the block devices, and use the change command to swap out a device. Press Ctrl+Alt+1 to go back to the virtual machine.

Running virtualized system

qemu-system-* binaries (for example qemu-system-i386 or qemu-system-x86_64, depending on guest's architecture) are used to run the virtualized guest. The usage is:

$ qemu-system-x86_64 options disk_image

Options are the same for all qemu-system-* binaries, see qemu(1) for documentation of all options.

By default, QEMU will show the virtual machine's video output in a window. One thing to keep in mind: when you click inside the QEMU window, the mouse pointer is grabbed. To release it, press Ctrl+Alt+g.

Warning: QEMU should never be run as root. If you must launch it in a script as root, you should use the -runas option to make QEMU drop root privileges.

Enabling KVM

KVM must be supported by your processor and kernel, and necessary kernel modules must be loaded. See KVM for more information.

To start QEMU in KVM mode, append -enable-kvm to the additional start options. To check if KVM is enabled for a running VM, enter the #QEMU monitor using Ctrl+Alt+Shift+2, and type info kvm.

Note:
  • The argument accel=kvm of the -machine option is equivalent to the -enable-kvm option.
  • If you start your VM with a GUI tool and experience very bad performance, you should check for proper KVM support, as QEMU may be falling back to software emulation.
  • KVM needs to be enabled in order to start Windows 7 and Windows 8 properly without a blue screen.

Enabling IOMMU (Intel VT-d/AMD-Vi) support

First enable IOMMU, see PCI passthrough via OVMF#Setting up IOMMU.

Add -device intel-iommu to create the IOMMU device:

$ qemu-system-x86_64 -enable-kvm -machine q35 -device intel-iommu -cpu host ..
Note: On Intel CPU based systems creating an IOMMU device in a QEMU guest with -device intel-iommu will disable PCI passthrough with an error like:
Device at bus pcie.0 addr 09.0 requires iommu notifier which is currently not supported by intel-iommu emulation
While adding the kernel parameter intel_iommu=on is still needed for remapping IO (e.g. PCI passthrough with vfio-pci), -device intel-iommu should not be set if PCI PCI passthrough is required.

Sharing data between host and guest

Network

Data can be shared between the host and guest OS using any network protocol that can transfer files, such as NFS, SMB, NBD, HTTP, FTP, or SSH, provided that you have set up the network appropriately and enabled the appropriate services.

The default user-mode networking allows the guest to access the host OS at the IP address 10.0.2.2. Any servers that you are running on your host OS, such as a SSH server or SMB server, will be accessible at this IP address. So on the guests, you can mount directories exported on the host via SMB or NFS, or you can access the host's HTTP server, etc. It will not be possible for the host OS to access servers running on the guest OS, but this can be done with other network configurations (see #Tap networking with QEMU).

QEMU's port forwarding

QEMU can forward ports from the host to the guest to enable e.g. connecting from the host to a SSH-server running on the guest.

For example, to bind port 10022 on the host with port 22 (SSH) on the guest, start QEMU with a command like:

$ qemu-system-x86_64 disk_image -net nic -net user,hostfwd=tcp::10022-:22

Make sure the sshd is running on the guest and connect with:

$ ssh guest-user@localhost -p10022

QEMU's built-in SMB server

QEMU's documentation says it has a "built-in" SMB server, but actually it just starts up Samba with an automatically generated smb.conf file located in /tmp/qemu-smb.random_string and makes it accessible to the guest at a different IP address (10.0.2.4 by default). This only works for user networking, and is useful when you don't want to start the normal Samba service on the host, which the guest can also access if you have set up shares on it.

Only a single directory can be set as shared with the option smb=, but adding more directories (even while the virtual machine is running) could be as easy as creating symbolic links in the shared directory if QEMU configured SMB to follow symbolic links. It doesn't do so, but the configuration of the running SMB server can be changed as described below.

To enable this feature, start QEMU with a command like:

$ qemu-system-x86_64 disk_image -net nic -net user,smb=shared_dir_path

where shared_dir_path is a directory that you want to share between the guest and host.

Then, in the guest, you will be able to access the shared directory on the host 10.0.2.4 with the share name "qemu". For example, in Windows Explorer you would go to \\10.0.2.4\qemu.

Note:
  • If you are using sharing options multiple times like -net user,smb=shared_dir_path1 -net user,smb=shared_dir_path2 or -net user,smb=shared_dir_path1,smb=shared_dir_path2 then it will share only the last defined one.
  • If you cannot access the shared folder and the guest system is Windows, check that the NetBIOS protocol is enabled and that a firewall does not block ports used by the NetBIOS protocol.
  • If you cannot access the shared folder and the guest system is Windows 10 Enterprise or Education or Windows Server 2016, enable guest access.

One way to share multiple directories and to add or remove them while the virtual machine is running, is to share an empty directory and create/remove symbolic links to the directories in the shared directory. For this to work, the configuration of the running SMB server can be changed with the following script, which also allows the execution of files on the guest that are not set executable on the host:

#!/bin/bash
eval $(ps h -C smbd -o pid,args | grep /tmp/qemu-smb | gawk '{print "pid="$1";conf="$6}')
echo "[global]
allow insecure wide links = yes
[qemu]
follow symlinks = yes
wide links = yes
acl allow execute always = yes" >> $conf
# in case the change is not detected automatically:
smbcontrol --configfile=$conf $pid reload-config

This can be applied to the running server started by qemu only after the guest has connected to the network drive the first time. An alternative to this method is to add additional shares to the configuration file like so:

echo "[myshare]
path=another_path
read only=no
guest ok=yes
force user=username" >> $conf

This share will be available on the guest as \\10.0.2.4\myshare.

Using filesystem passthrough and VirtFS

See the QEMU documentation.

Mounting a partition of the guest on the host

It can be useful to mount a drive image under the host system, it can be a way to transfer files in and out of the guest. This should be done when the virtual machine is not running.

The procedure to mount the drive on the host depends on the type of qemu image, raw or qcow2. We detail thereafter the steps to mount a drive in the two formats in #Mounting a partition from a raw image and #Mounting a partition from a qcow2 image. For the full docummentation see QEMU/Mounting an image on the host.

Warning: You must make sure to unmount the partitions before running the virtual machine again. Otherwise, data corruption is very likely to occur.

Mounting a partition from a raw image

It is possible to mount partitions that are inside a raw disk image file by setting them up as loopback devices.

With manually specifying byte offset

One way to mount a disk image partition is to mount the disk image at a certain offset using a command like the following:

# mount -o loop,offset=32256 disk_image mountpoint

The offset=32256 option is actually passed to the losetup program to set up a loopback device that starts at byte offset 32256 of the file and continues to the end. This loopback device is then mounted. You may also use the sizelimit option to specify the exact size of the partition, but this is usually unnecessary.

Depending on your disk image, the needed partition may not start at offset 32256. Run fdisk -l disk_image to see the partitions in the image. fdisk gives the start and end offsets in 512-byte sectors, so multiply by 512 to get the correct offset to pass to mount.

With loop module autodetecting partitions

The Linux loop driver actually supports partitions in loopback devices, but it is disabled by default. To enable it, do the following:

  • Get rid of all your loopback devices (unmount all mounted images, etc.).
  • Unload the loop kernel module, and load it with the max_part=15 parameter set. Additionally, the maximum number of loop devices can be controlled with the max_loop parameter.
Tip: You can put an entry in /etc/modprobe.d to load the loop module with max_part=15 every time, or you can put loop.max_part=15 on the kernel command-line, depending on whether you have the loop.ko module built into your kernel or not.

Set up your image as a loopback device:

# losetup -f -P disk_image

Then, if the device created was /dev/loop0, additional devices /dev/loop0pX will have been automatically created, where X is the number of the partition. These partition loopback devices can be mounted directly. For example:

# mount /dev/loop0p1 mountpoint

To mount the disk image with udisksctl, see Udisks#Mount loop devices.

With kpartx

kpartx from the multipath-tools package can read a partition table on a device and create a new device for each partition. For example:

# kpartx -a disk_image

This will setup the loopback device and create the necessary partition(s) device(s) in /dev/mapper/.

Mounting a partition from a qcow2 image

We will use qemu-nbd, which lets use the NBD (network block device) protocol to share the disk image.

First, we need the nbd module loaded:

# modprobe nbd max_part=16

Then, we can share the disk and create the device entries:

# qemu-nbd -c /dev/nbd0 /path/to/image.qcow2

Discover the partitions:

# partprobe /dev/nbd0

fdisk can be used to get information regarding the different partitions in nbd0:

# fdisk -l /dev/nbd0
Disk /dev/nbd0: 25.2 GiB, 27074281472 bytes, 52879456 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0xa6a4d542

Device      Boot   Start      End  Sectors  Size Id Type
/dev/nbd0p1 *       2048  1026047  1024000  500M  7 HPFS/NTFS/exFAT
/dev/nbd0p2      1026048 52877311 51851264 24.7G  7 HPFS/NTFS/exFAT

Then mount any partition of the drive image, for example the partition 2:

# mount /dev/nbd0p2 mountpoint

After the usage, it is important to unmount the image and reverse previous steps, i.e. unmount the partition and disconnect the nbd device:

  1. umount mountpoint
  2. qemu-nbd -d /dev/nbd0

Using any real partition as the single primary partition of a hard disk image

Sometimes, you may wish to use one of your system partitions from within QEMU. Using a raw partition for a virtual machine will improve performance, as the read and write operations do not go through the file system layer on the physical host. Such a partition also provides a way to share data between the host and guest.

In Arch Linux, device files for raw partitions are, by default, owned by root and the disk group. If you would like to have a non-root user be able to read and write to a raw partition, you need to change the owner of the partition's device file to that user.

Warning:
  • Although it is possible, it is not recommended to allow virtual machines to alter critical data on the host system, such as the root partition.
  • You must not mount a file system on a partition read-write on both the host and the guest at the same time. Otherwise, data corruption will result.

After doing so, you can attach the partition to a QEMU virtual machine as a virtual disk.

However, things are a little more complicated if you want to have the entire virtual machine contained in a partition. In that case, there would be no disk image file to actually boot the virtual machine since you cannot install a bootloader to a partition that is itself formatted as a file system and not as a partitioned device with an MBR. Such a virtual machine can be booted either by specifying the kernel and initrd manually, or by simulating a disk with an MBR by using the Device-mapper, linear RAID, or a Linux Network Block Device.

By specifying kernel and initrd manually

QEMU supports loading Linux kernels and init ramdisks directly, thereby circumventing bootloaders such as GRUB. It then can be launched with the physical partition containing the root file system as the virtual disk, which will not appear to be partitioned. This is done by issuing a command similar to the following:

Note: In this example, it is the host's images that are being used, not the guest's. If you wish to use the guest's images, either mount /dev/sda3 read-only (to protect the file system from the host) and specify the /full/path/to/images or use some kexec hackery in the guest to reload the guest's kernel (extends boot time).
$ qemu-system-x86_64 -kernel /boot/vmlinuz-linux -initrd /boot/initramfs-linux.img -append root=/dev/sda /dev/sda3

In the above example, the physical partition being used for the guest's root file system is /dev/sda3 on the host, but it shows up as /dev/sda on the guest.

You may, of course, specify any kernel and initrd that you want, and not just the ones that come with Arch Linux.

When there are multiple kernel parameters to be passed to the -append option, they need to be quoted using single or double quotes. For example:

... -append 'root=/dev/sda1 console=ttyS0'

Simulate a virtual disk with MBR

A more complicated way to have a virtual machine use a physical partition, while keeping that partition formatted as a file system and not just having the guest partition the partition as if it were a disk, is to simulate an MBR for it so that it can boot using a bootloader such as GRUB.

For the following, suppose you have a plain, unmounted /dev/hdaN partition with some file system on it you wish to make part of a QEMU disk image. The trick is to dynamically prepend a master boot record (MBR) to the real partition you wish to embed in a QEMU raw disk image. More generally, the partition can be any part of a larger simulated disk, in particular a block device that simulates the original physical disk but only exposes /dev/hdaN to the virtual machine.

A virtual disk of this type can be represented by a VMDK file that contains references to (a copy of) the MBR and the partition, but QEMU does not support this VMDK format. For instance, a virtual disk created by

$ VBoxManage internalcommands createrawvmdk -filename /path/to/file.vmdk -rawdisk /dev/hda

will be rejected by QEMU with the error message

Unsupported image type 'partitionedDevice'

Note that VBoxManage creates two files, file.vmdk and file-pt.vmdk, the latter being a copy of the MBR, to which the text file file.vmdk points. Read operations outside the target partition or the MBR would give zeros, while written data would be discarded.

Device Mapper

A method that is similar to the use of a VMDK descriptor file uses the device mapper to prepend a loop device attached to the MBR file to the target partition. In case we don't need our virtual disk to have the same size as the original, we first create a file to hold the MBR:

$ dd if=/dev/zero of=/path/to/mbr count=2048

Here, a 1 MB (2048 * 512 bytes) file is created in accordance with partition alignment policies used by modern disk partitioning tools. For compatibility with older partitioning software, 63 sectors instead of 2048 might be required. The MBR only needs a single 512 bytes block, the additional free space can be used for a BIOS boot partition and, in the case of a hybrid partitioning scheme, for a GUID Partition Table. Then, we attach a loop device to the MBR file:

# losetup --show -f /path/to/mbr
/dev/loop0

In this example, the resulting device is /dev/loop0. The device mapper is now used to join the MBR and the partition:

# echo "0 2048 linear /dev/loop0 0
2048 `blockdev --getsz /dev/hdaN` linear /dev/hdaN 0" | dmsetup create qemu

The resulting /dev/mapper/qemu is what we will use as a QEMU raw disk image. Additional steps are required to create a partition table (see the section that describes the use of a linear RAID for an example) and boot loader code on the virtual disk (which will be stored in /path/to/mbr).

The following setup is an example where the position of /dev/hdaN on the virtual disk is to be the same as on the physical disk and the rest of the disk is hidden, except for the MBR, which is provided as a copy:

# dd if=/dev/hda count=1 of=/path/to/mbr
# loop=`losetup --show -f /path/to/mbr`
# start=`blockdev --report /dev/hdaN | tail -1 | awk '{print $5}'`
# size=`blockdev --getsz /dev/hdaN`
# disksize=`blockdev --getsz /dev/hda`
# echo "0 1 linear $loop 0
1 $((start-1)) zero
$start $size linear /dev/hdaN 0
$((start+size)) $((disksize-start-size)) zero" | dmsetup create qemu

The table provided as standard input to dmsetup has a similar format as the table in a VDMK descriptor file produced by VBoxManage and can alternatively be loaded from a file with dmsetup create qemu --table table_file. To the virtual machine, only /dev/hdaN is accessible, while the rest of the hard disk reads as zeros and discards written data, except for the first sector. We can print the table for /dev/mapper/qemu with dmsetup table qemu (use udevadm info -rq name /sys/dev/block/major:minor to translate major:minor to the corresponding /dev/blockdevice name). Use dmsetup remove qemu and losetup -d $loop to delete the created devices.

A situation where this example would be useful is an existing Windows XP installation in a multi-boot configuration and maybe a hybrid partitioning scheme (on the physical hardware, Windows XP could be the only operating system that uses the MBR partition table, while more modern operating systems installed on the same computer could use the GUID Partition Table). Windows XP supports hardware profiles, so that that the same installation can be used with different hardware configurations alternatingly (in this case bare metal vs. virtual) with Windows needing to install drivers for newly detected hardware only once for every profile. Note that in this example the boot loader code in the copied MBR needs to be updated to directly load Windows XP from /dev/hdaN instead of trying to start the multi-boot capable boot loader (like GRUB) present in the original system. Alternatively, a copy of the boot partition containing the boot loader installation can be included in the virtual disk the same way as the MBR.

Linear RAID

You can also do this using software RAID in linear mode (you need the linear.ko kernel driver) and a loopback device:

First, you create some small file to hold the MBR:

$ dd if=/dev/zero of=/path/to/mbr count=32

Here, a 16 KB (32 * 512 bytes) file is created. It is important not to make it too small (even if the MBR only needs a single 512 bytes block), since the smaller it will be, the smaller the chunk size of the software RAID device will have to be, which could have an impact on performance. Then, you setup a loopback device to the MBR file:

# losetup -f /path/to/mbr

Let us assume the resulting device is /dev/loop0, because we would not already have been using other loopbacks. Next step is to create the "merged" MBR + /dev/hdaN disk image using software RAID:

# modprobe linear
# mdadm --build --verbose /dev/md0 --chunk=16 --level=linear --raid-devices=2 /dev/loop0 /dev/hdaN

The resulting /dev/md0 is what you will use as a QEMU raw disk image (do not forget to set the permissions so that the emulator can access it). The last (and somewhat tricky) step is to set the disk configuration (disk geometry and partitions table) so that the primary partition start point in the MBR matches the one of /dev/hdaN inside /dev/md0 (an offset of exactly 16 * 512 = 16384 bytes in this example). Do this using fdisk on the host machine, not in the emulator: the default raw disc detection routine from QEMU often results in non-kilobyte-roundable offsets (such as 31.5 KB, as in the previous section) that cannot be managed by the software RAID code. Hence, from the the host:

# fdisk /dev/md0

Press X to enter the expert menu. Set number of 's'ectors per track so that the size of one cylinder matches the size of your MBR file. For two heads and a sector size of 512, the number of sectors per track should be 16, so we get cylinders of size 2x16x512=16k.

Now, press R to return to the main menu.

Press P and check that the cylinder size is now 16k.

Now, create a single primary partition corresponding to /dev/hdaN. It should start at cylinder 2 and end at the end of the disk (note that the number of cylinders now differs from what it was when you entered fdisk.

Finally, 'w'rite the result to the file: you are done. You now have a partition you can mount directly from your host, as well as part of a QEMU disk image:

$ qemu-system-x86_64 -hdc /dev/md0 [...]

You can, of course, safely set any bootloader on this disk image using QEMU, provided the original /dev/hdaN partition contains the necessary tools.

Network Block Device

Instead of the methods decribed above, you may use nbd-server (from the nbd package) to create an MBR wrapper for QEMU.

Assuming you have already set up your MBR wrapper file like above, rename it to wrapper.img.0. Then create a symbolic link named wrapper.img.1 in the same directory, pointing to your partition. Then put the following script in the same directory:

#!/bin/sh
dir="$(realpath "$(dirname "$0")")"
cat >wrapper.conf <<EOF
[generic]
allowlist = true
listenaddr = 127.713705
port = 10809

[wrap]
exportname = $dir/wrapper.img
multifile = true
EOF

nbd-server \
    -C wrapper.conf \
    -p wrapper.pid \
    "$@"

The .0 and .1 suffixes are essential; the rest can be changed. After running the above script (which you may need to do as root to make sure nbd-server is able to access the partition), you can launch QEMU with:

qemu-system-x86_64 -drive file=nbd:127.713705:10809:exportname=wrap [...]

Networking

Tango-edit-clear.pngThis article or section needs language, wiki syntax or style improvements. See Help:Style for reference.Tango-edit-clear.png

Reason: Network topologies (sections #Host-only networking, #Internal networking and info spread out across other sections) should not be described alongside the various virtual interfaces implementations, such as #User-mode networking, #Tap networking with QEMU, #Networking with VDE2. (Discuss in Talk:QEMU#)

The performance of virtual networking should be better with tap devices and bridges than with user-mode networking or vde because tap devices and bridges are implemented in-kernel.

In addition, networking performance can be improved by assigning virtual machines a virtio network device rather than the default emulation of an e1000 NIC. See #Installing virtio drivers for more information.

Link-level address caveat

By giving the -net nic argument to QEMU, it will, by default, assign a virtual machine a network interface with the link-level address 52:54:00:12:34:56. However, when using bridged networking with multiple virtual machines, it is essential that each virtual machine has a unique link-level (MAC) address on the virtual machine side of the tap device. Otherwise, the bridge will not work correctly, because it will receive packets from multiple sources that have the same link-level address. This problem occurs even if the tap devices themselves have unique link-level addresses because the source link-level address is not rewritten as packets pass through the tap device.

Make sure that each virtual machine has a unique link-level address, but it should always start with 52:54:. Use the following option, replace X with arbitrary hexadecimal digit:

$ qemu-system-x86_64 -net nic,macaddr=52:54:XX:XX:XX:XX -net vde disk_image

Generating unique link-level addresses can be done in several ways:

  1. Manually specify unique link-level address for each NIC. The benefit is that the DHCP server will assign the same IP address each time the virtual machine is run, but it is unusable for large number of virtual machines.
  2. Generate random link-level address each time the virtual machine is run. Practically zero probability of collisions, but the downside is that the DHCP server will assign a different IP address each time. You can use the following command in a script to generate random link-level address in a macaddr variable:
    printf -v macaddr "52:54:%02x:%02x:%02x:%02x" $(( $RANDOM & 0xff)) $(( $RANDOM & 0xff )) $(( $RANDOM & 0xff)) $(( $RANDOM & 0xff ))
    qemu-system-x86_64 -net nic,macaddr="$macaddr" -net vde disk_image
  3. Use the following script qemu-mac-hasher.py to generate the link-level address from the virtual machine name using a hashing function. Given that the names of virtual machines are unique, this method combines the benefits of the aforementioned methods: it generates the same link-level address each time the script is run, yet it preserves the practically zero probability of collisions.
    qemu-mac-hasher.py
    #!/usr/bin/env python
    
    import sys
    import zlib
    
    if len(sys.argv) != 2:
        print("usage: %s <VM Name>" % sys.argv[0])
        sys.exit(1)
    
    crc = zlib.crc32(sys.argv[1].encode("utf-8")) & 0xffffffff
    crc = str(hex(crc))[2:]
    print("52:54:%s%s:%s%s:%s%s:%s%s" % tuple(crc))
    

    In a script, you can use for example:

    vm_name="VM Name"
    qemu-system-x86_64 -name "$vm_name" -net nic,macaddr=$(qemu-mac-hasher.py "$vm_name") -net vde disk_image
    

User-mode networking

By default, without any -netdev arguments, QEMU will use user-mode networking with a built-in DHCP server. Your virtual machines will be assigned an IP address when they run their DHCP client, and they will be able to access the physical host's network through IP masquerading done by QEMU.

Warning: This only works with the TCP and UDP protocols, so ICMP, including ping, will not work. Do not use ping to test network connectivity. To make ping work in the guest see Sysctl#Allow_unprivileged_users_to_create_IPPROTO_ICMP_sockets.

This default configuration allows your virtual machines to easily access the Internet, provided that the host is connected to it, but the virtual machines will not be directly visible on the external network, nor will virtual machines be able to talk to each other if you start up more than one concurrently.

QEMU's user-mode networking can offer more capabilities such as built-in TFTP or SMB servers, redirecting host ports to the guest (for example to allow SSH connections to the guest) or attaching guests to VLANs so that they can talk to each other. See the QEMU documentation on the -net user flag for more details.

However, user-mode networking has limitations in both utility and performance. More advanced network configurations require the use of tap devices or other methods.

Note: If the host system uses systemd-networkd, make sure to symlink the /etc/resolv.conf file as described in systemd-networkd#Required services and setup, otherwise the DNS lookup in the guest system will not work.

Tap networking with QEMU

Tap devices are a Linux kernel feature that allows you to create virtual network interfaces that appear as real network interfaces. Packets sent to a tap interface are delivered to a userspace program, such as QEMU, that has bound itself to the interface.

QEMU can use tap networking for a virtual machine so that packets sent to the tap interface will be sent to the virtual machine and appear as coming from a network interface (usually an Ethernet interface) in the virtual machine. Conversely, everything that the virtual machine sends through its network interface will appear on the tap interface.

Tap devices are supported by the Linux bridge drivers, so it is possible to bridge together tap devices with each other and possibly with other host interfaces such as eth0. This is desirable if you want your virtual machines to be able to talk to each other, or if you want other machines on your LAN to be able to talk to the virtual machines.

Warning: If you bridge together tap device and some host interface, such as eth0, your virtual machines will appear directly on the external network, which will expose them to possible attack. Depending on what resources your virtual machines have access to, you may need to take all the precautions you normally would take in securing a computer to secure your virtual machines. If the risk is too great, virtual machines have little resources or you set up multiple virtual machines, a better solution might be to use host-only networking and set up NAT. In this case you only need one firewall on the host instead of multiple firewalls for each guest.

As indicated in the user-mode networking section, tap devices offer higher networking performance than user-mode. If the guest OS supports virtio network driver, then the networking performance will be increased considerably as well. Supposing the use of the tap0 device, that the virtio driver is used on the guest, and that no scripts are used to help start/stop networking, next is part of the qemu command one should see:

-device virtio-net,netdev=network0 -netdev tap,id=network0,ifname=tap0,script=no,downscript=no

But if already using a tap device with virtio networking driver, one can even boost the networking performance by enabling vhost, like:

-device virtio-net,netdev=network0 -netdev tap,id=network0,ifname=tap0,script=no,downscript=no,vhost=on

See http://www.linux-kvm.com/content/how-maximize-virtio-net-performance-vhost-net for more information.

Host-only networking

If the bridge is given an IP address and traffic destined for it is allowed, but no real interface (e.g. eth0) is connected to the bridge, then the virtual machines will be able to talk to each other and the host system. However, they will not be able to talk to anything on the external network, provided that you do not set up IP masquerading on the physical host. This configuration is called host-only networking by other virtualization software such as VirtualBox.

Tip:
  • If you want to set up IP masquerading, e.g. NAT for virtual machines, see the Internet sharing#Enable NAT page.
  • See Network bridge for information on creating bridge.
  • You may want to have a DHCP server running on the bridge interface to service the virtual network. For example, to use the 172.20.0.1/16 subnet with dnsmasq as the DHCP server:
# ip addr add 172.20.0.1/16 dev br0
# ip link set br0 up
# dnsmasq --interface=br0 --bind-interfaces --dhcp-range=172.20.0.2,172.20.255.254

Internal networking

If you do not give the bridge an IP address and add an iptables rule to drop all traffic to the bridge in the INPUT chain, then the virtual machines will be able to talk to each other, but not to the physical host or to the outside network. This configuration is called internal networking by other virtualization software such as VirtualBox. You will need to either assign static IP addresses to the virtual machines or run a DHCP server on one of them.

By default iptables would drop packets in the bridge network. You may need to use such iptables rule to allow packets in a bridged network:

# iptables -I FORWARD -m physdev --physdev-is-bridged -j ACCEPT

Bridged networking using qemu-bridge-helper

Note: This method is available since QEMU 1.1, see http://wiki.qemu.org/Features/HelperNetworking.

This method does not require a start-up script and readily accommodates multiple taps and multiple bridges. It uses /usr/lib/qemu/qemu-bridge-helper binary, which allows creating tap devices on an existing bridge.

Tip: See Network bridge for information on creating bridge.

First, create a configuration file containing the names of all bridges to be used by QEMU:

/etc/qemu/bridge.conf
allow bridge0
allow bridge1
...

Now start the VM. The most basic usage would be:

$ qemu-system-x86_64 -net nic -net bridge,br=bridge0 [...]

With multiple taps, the most basic usage requires specifying the VLAN for all additional NICs:

$ qemu-system-x86_64 -net nic -net bridge,br=bridge0 -net nic,vlan=1 -net bridge,vlan=1,br=bridge1 [...]

Creating bridge manually

Tango-edit-clear.pngThis article or section needs language, wiki syntax or style improvements. See Help:Style for reference.Tango-edit-clear.png

Reason: This section needs serious cleanup and may contain out-of-date information. (Discuss in Talk:QEMU#)
Tip: Since QEMU 1.1, the network bridge helper can set tun/tap up for you without the need for additional scripting. See #Bridged networking using qemu-bridge-helper.

The following describes how to bridge a virtual machine to a host interface such as eth0, which is probably the most common configuration. This configuration makes it appear that the virtual machine is located directly on the external network, on the same Ethernet segment as the physical host machine.

We will replace the normal Ethernet adapter with a bridge adapter and bind the normal Ethernet adapter to it.

  • Install bridge-utils, which provides brctl to manipulate bridges.
  • Enable IPv4 forwarding:
# sysctl net.ipv4.ip_forward=1

To make the change permanent, change net.ipv4.ip_forward = 0 to net.ipv4.ip_forward = 1 in /etc/sysctl.d/99-sysctl.conf.

  • Load the tun module and configure it to be loaded on boot. See Kernel modules for details.
  • Now create the bridge. See Bridge with netctl for details. Remember to name your bridge as br0, or change the scripts below to your bridge's name.
  • Create the script that QEMU uses to bring up the tap adapter with root:kvm 750 permissions:
/etc/qemu-ifup
#!/bin/sh

echo "Executing /etc/qemu-ifup"
echo "Bringing up $1 for bridged mode..."
sudo /usr/bin/ip link set $1 up promisc on
echo "Adding $1 to br0..."
sudo /usr/bin/brctl addif br0 $1
sleep 2
  • Create the script that QEMU uses to bring down the tap adapter in /etc/qemu-ifdown with root:kvm 750 permissions:
/etc/qemu-ifdown
#!/bin/sh

echo "Executing /etc/qemu-ifdown"
sudo /usr/bin/ip link set $1 down
sudo /usr/bin/brctl delif br0 $1
sudo /usr/bin/ip link delete dev $1
  • Use visudo to add the following to your sudoers file:
Cmnd_Alias      QEMU=/usr/bin/ip,/usr/bin/modprobe,/usr/bin/brctl
%kvm     ALL=NOPASSWD: QEMU
  • You launch QEMU using the following run-qemu script:
run-qemu
#!/bin/bash
USERID=$(whoami)

# Get name of newly created TAP device; see https://bbs.archlinux.org/viewtopic.php?pid=1285079#p1285079
precreationg=$(/usr/bin/ip tuntap list | /usr/bin/cut -d: -f1 | /usr/bin/sort)
sudo /usr/bin/ip tuntap add user $USERID mode tap
postcreation=$(/usr/bin/ip tuntap list | /usr/bin/cut -d: -f1 | /usr/bin/sort)
IFACE=$(comm -13 <(echo "$precreationg") <(echo "$postcreation"))

# This line creates a random MAC address. The downside is the DHCP server will assign a different IP address each time
printf -v macaddr "52:54:%02x:%02x:%02x:%02x" $(( $RANDOM & 0xff)) $(( $RANDOM & 0xff )) $(( $RANDOM & 0xff)) $(( $RANDOM & 0xff ))
# Instead, uncomment and edit this line to set a static MAC address. The benefit is that the DHCP server will assign the same IP address.
# macaddr='52:54:be:36:42:a9'

qemu-system-x86_64 -net nic,macaddr=$macaddr -net tap,ifname="$IFACE" $*

sudo ip link set dev $IFACE down &> /dev/null
sudo ip tuntap del $IFACE mode tap &> /dev/null

Then to launch a VM, do something like this

$ run-qemu -hda myvm.img -m 512
/etc/sysctl.d/10-disable-firewall-on-bridge.conf
net.bridge.bridge-nf-call-ip6tables = 0
net.bridge.bridge-nf-call-iptables = 0
net.bridge.bridge-nf-call-arptables = 0

Run sysctl -p /etc/sysctl.d/10-disable-firewall-on-bridge.conf to apply the changes immediately.

See the libvirt wiki and Fedora bug 512206. If you get errors by sysctl during boot about non-existing files, make the bridge module load at boot. See Kernel modules#Automatic module loading with systemd.

Alternatively, you can configure iptables to allow all traffic to be forwarded across the bridge by adding a rule like this:

-I FORWARD -m physdev --physdev-is-bridged -j ACCEPT

Network sharing between physical device and a Tap device through iptables

Merge-arrows-2.pngThis article or section is a candidate for merging with Internet_sharing.Merge-arrows-2.png

Notes: Duplication, not specific to QEMU. (Discuss in Talk:QEMU#)

Bridged networking works fine between a wired interface (Eg. eth0), and it is easy to setup. However if the host gets connected to the network through a wireless device, then bridging is not possible.

See Network bridge#Wireless interface on a bridge as a reference.

One way to overcome that is to setup a tap device with a static IP, making linux automatically handle the routing for it, and then forward traffic between the tap interface and the device connected to the network through iptables rules.

See Internet sharing as a reference.

There you can find what is needed to share the network between devices, included tap and tun ones. The following just hints further on some of the host configurations required. As indicated in the reference above, the client needs to be configured for a static IP, using the IP assigned to the tap interface as the gateway. The caveat is that the DNS servers on the client might need to be manually edited if they change when changing from one host device connected to the network to another.

To allow IP forwarding on every boot, one need to add the following lines to sysctl configuration file inside /etc/sysctl.d:

net.ipv4.ip_forward = 1
net.ipv6.conf.default.forwarding = 1
net.ipv6.conf.all.forwarding = 1

The iptables rules can look like:

# Forwarding from/to outside
iptables -A FORWARD -i ${INT} -o ${EXT_0} -j ACCEPT
iptables -A FORWARD -i ${INT} -o ${EXT_1} -j ACCEPT
iptables -A FORWARD -i ${INT} -o ${EXT_2} -j ACCEPT
iptables -A FORWARD -i ${EXT_0} -o ${INT} -j ACCEPT
iptables -A FORWARD -i ${EXT_1} -o ${INT} -j ACCEPT
iptables -A FORWARD -i ${EXT_2} -o ${INT} -j ACCEPT
# NAT/Masquerade (network address translation)
iptables -t nat -A POSTROUTING -o ${EXT_0} -j MASQUERADE
iptables -t nat -A POSTROUTING -o ${EXT_1} -j MASQUERADE
iptables -t nat -A POSTROUTING -o ${EXT_2} -j MASQUERADE

The prior supposes there are 3 devices connected to the network sharing traffic with one internal device, where for example:

INT=tap0
EXT_0=eth0
EXT_1=wlan0
EXT_2=tun0

The prior shows a forwarding that would allow sharing wired and wireless connections with the tap device.

The forwarding rules shown are stateless, and for pure forwarding. One could think of restricting specific traffic, putting a firewall in place to protect the guest and others. However those would decrease the networking performance, while a simple bridge does not include any of that.

Bonus: Whether the connection is wired or wireless, if one gets connected through VPN to a remote site with a tun device, supposing the tun device opened for that connection is tun0, and the prior iptables rules are applied, then the remote connection gets also shared with the guest. This avoids the need for the guest to also open a VPN connection. Again, as the guest networking needs to be static, then if connecting the host remotely this way, one most probably will need to edit the DNS servers on the guest.

Networking with VDE2

Tango-edit-clear.pngThis article or section needs language, wiki syntax or style improvements. See Help:Style for reference.Tango-edit-clear.png

Reason: This section needs serious cleanup and may contain out-of-date information. (Discuss in Talk:QEMU#)

What is VDE?

VDE stands for Virtual Distributed Ethernet. It started as an enhancement of uml_switch. It is a toolbox to manage virtual networks.

The idea is to create virtual switches, which are basically sockets, and to "plug" both physical and virtual machines in them. The configuration we show here is quite simple; However, VDE is much more powerful than this, it can plug virtual switches together, run them on different hosts and monitor the traffic in the switches. You are invited to read the documentation of the project.

The advantage of this method is you do not have to add sudo privileges to your users. Regular users should not be allowed to run modprobe.

Basics

VDE support can be installed via the vde2 package.

In our config, we use tun/tap to create a virtual interface on my host. Load the tun module (see Kernel modules for details):

# modprobe tun

Now create the virtual switch:

# vde_switch -tap tap0 -daemon -mod 660 -group users

This line creates the switch, creates tap0, "plugs" it, and allows the users of the group users to use it.

The interface is plugged in but not configured yet. To configure it, run this command:

# ip addr add 192.168.100.254/24 dev tap0

Now, you just have to run KVM with these -net options as a normal user:

$ qemu-system-x86_64 -net nic -net vde -hda [...]

Configure networking for your guest as you would do in a physical network.

Tip: You might want to set up NAT on tap device to access the internet from the virtual machine. See Internet sharing#Enable NAT for more information.

Startup scripts

Example of main script starting VDE:

/etc/systemd/scripts/qemu-network-env
#!/bin/sh
# QEMU/VDE network environment preparation script

# The IP configuration for the tap device that will be used for
# the virtual machine network:

TAP_DEV=tap0
TAP_IP=192.168.100.254
TAP_MASK=24
TAP_NETWORK=192.168.100.0

# Host interface
NIC=eth0

case "$1" in
  start)
        echo -n "Starting VDE network for QEMU: "

        # If you want tun kernel module to be loaded by script uncomment here
	#modprobe tun 2>/dev/null
	## Wait for the module to be loaded
 	#while ! lsmod | grep -q "^tun"; do echo "Waiting for tun device"; sleep 1; done

        # Start tap switch
        vde_switch -tap "$TAP_DEV" -daemon -mod 660 -group users

        # Bring tap interface up
        ip address add "$TAP_IP"/"$TAP_MASK" dev "$TAP_DEV"
        ip link set "$TAP_DEV" up

        # Start IP Forwarding
        echo "1" > /proc/sys/net/ipv4/ip_forward
        iptables -t nat -A POSTROUTING -s "$TAP_NETWORK"/"$TAP_MASK" -o "$NIC" -j MASQUERADE
        ;;
  stop)
        echo -n "Stopping VDE network for QEMU: "
        # Delete the NAT rules
        iptables -t nat -D POSTROUTING -s "$TAP_NETWORK"/"$TAP_MASK" -o "$NIC" -j MASQUERADE

        # Bring tap interface down
        ip link set "$TAP_DEV" down

        # Kill VDE switch
        pgrep vde_switch | xargs kill -TERM
        ;;
  restart|reload)
        $0 stop
        sleep 1
        $0 start
        ;;
  *)
        echo "Usage: $0 {start|stop|restart|reload}"
        exit 1
esac
exit 0

Example of systemd service using the above script:

/etc/systemd/system/qemu-network-env.service
[Unit]
Description=Manage VDE Switch

[Service]
Type=oneshot
ExecStart=/etc/systemd/scripts/qemu-network-env start
ExecStop=/etc/systemd/scripts/qemu-network-env stop
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target

Change permissions for qemu-network-env to be executable

# chmod u+x /etc/systemd/scripts/qemu-network-env

You can start qemu-network-env.service as usual.

Alternative method

If the above method does not work or you do not want to mess with kernel configs, TUN, dnsmasq, and iptables you can do the following for the same result.

# vde_switch -daemon -mod 660 -group users
# slirpvde --dhcp --daemon

Then, to start the VM with a connection to the network of the host:

$ qemu-system-x86_64 -net nic,macaddr=52:54:00:00:EE:03 -net vde disk_image

VDE2 Bridge

Based on quickhowto: qemu networking using vde, tun/tap, and bridge graphic. Any virtual machine connected to vde is externally exposed. For example, each virtual machine can receive DHCP configuration directly from your ADSL router.

Basics

Remember that you need tun module and bridge-utils package.

Create the vde2/tap device:

# vde_switch -tap tap0 -daemon -mod 660 -group users
# ip link set tap0 up

Create bridge:

# brctl addbr br0

Add devices:

# brctl addif br0 eth0
# brctl addif br0 tap0

And configure bridge interface:

# dhcpcd br0

Startup scripts

All devices must be set up. And only the bridge needs an IP address. For physical devices on the bridge (e.g. eth0), this can be done with netctl using a custom Ethernet profile with:

/etc/netctl/ethernet-noip
Description='A more versatile static Ethernet connection'
Interface=eth0
Connection=ethernet
IP=no

The following custom systemd service can be used to create and activate a VDE2 tap interface for use in the users user group.

/etc/systemd/system/vde2@.service
[Unit]
Description=Network Connectivity for %i
Wants=network.target
Before=network.target

[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/usr/bin/vde_switch -tap %i -daemon -mod 660 -group users
ExecStart=/usr/bin/ip link set dev %i up
ExecStop=/usr/bin/ip addr flush dev %i
ExecStop=/usr/bin/ip link set dev %i down

[Install]
WantedBy=multi-user.target

And finally, you can create the bridge interface with netctl.

Shorthand configuration

If you're using QEMU with various networking options a lot, you probably have created a lot of -netdev and -device argument pairs, which gets quite repetitive. You can instead use the -nic argument to combine -netdev and -device together, so that, for example, these arguments:

-netdev tap,id=network0,ifname=tap0,script=no,downscript=no,vhost=on -device virtio-net,netdev=network0

...become:

-nic tap,ifname=tap0,script=no,downscript=no,vhost=on,model=virtio-net

Notice the lack of network IDs, and that the device was created with model=.... The first half of the -nic parameters are -netdev parameters, whereas the second half (after model=...) are related with the device. The same parameters (for example, smb=...) are used. There's also a special parameter for -nic which completely disables the default (user-mode) networking:

-nic none

See QEMU networking documentation for more information on parameters you can use.

Graphic card

QEMU can emulate a standard graphic card text mode using -curses command line option. This allows to type text and see text output directly inside a text terminal.

QEMU can emulate several types of VGA card. The card type is passed in the -vga type command line option and can be std, qxl, vmware, virtio, cirrus or none.

std

With -vga std you can get a resolution of up to 2560 x 1600 pixels without requiring guest drivers. This is the default since QEMU 2.2.

qxl

QXL is a paravirtual graphics driver with 2D support. To use it, pass the -vga qxl option and install drivers in the guest. You may want to use #SPICE for improved graphical performance when using QXL.

On Linux guests, the qxl and bochs_drm kernel modules must be loaded in order to gain a decent performance.

Default VGA memory size for QXL devices is 16M which is sufficient to drive resolutions approximately up to QHD (2560x1440). To enable higher resolutions, increase vga_memmb.

vmware

Although it is a bit buggy, it performs better than std and cirrus. Install the VMware drivers xf86-video-vmware and xf86-input-vmmouse for Arch Linux guests.

virtio

virtio-vga / virtio-gpu is a paravirtual 3D graphics driver based on virgl. Currently a work in progress, supporting only very recent (>= 4.4) Linux guests with mesa (>=11.2) compiled with the option gallium-drivers=virgl.

To enable 3D acceleration on the guest system select this vga with -vga virtio and enable the opengl context in the display device with -display sdl,gl=on or -display gtk,gl=on for the sdl and gtk display output respectively. Successful configuration can be confirmed looking at the kernel log in the guest:

$ dmesg | grep drm 
[drm] pci: virtio-vga detected
[drm] virgl 3d acceleration enabled

cirrus

The cirrus graphical adapter was the default before 2.2. It should not be used on modern systems.

none

This is like a PC that has no VGA card at all. You would not even be able to access it with the -vnc option. Also, this is different from the -nographic option which lets QEMU emulate a VGA card, but disables the SDL display.

SPICE

The SPICE project aims to provide a complete open source solution for remote access to virtual machines in a seamless way. SPICE can only be used when using #qxl as the graphical output.

Enabling SPICE via the command line

The following is an example of booting with SPICE as the remote desktop protocol, including the support for copy and paste from host:

$ qemu-system-x86_64 -vga qxl -device virtio-serial-pci -spice port=5930,disable-ticketing -device virtserialport,chardev=spicechannel0,name=com.redhat.spice.0 -chardev spicevmc,id=spicechannel0,name=vdagent

The parameters have the following meaning:

  1. -device virtio-serial-pci adds a virtio-serial device
  2. -spice port=5930,disable-ticketing set TCP port 5930 for spice channels listening and allow client to connect without authentication
  3. -device virtserialport,chardev=spicechannel0,name=com.redhat.spice.0 opens a port for spice vdagent in the virtio-serial device,
  4. -chardev spicevmc,id=spicechannel0,name=vdagent adds a spicevmc chardev for that port. It is important that the chardev= option of the virtserialport device matches the id= option given to the chardev option (spicechannel0 in this example). It is also important that the port name is com.redhat.spice.0, because that is the namespace where vdagent is looking for in the guest. And finally, specify name=vdagent so that spice knows what this channel is for.
Tip:
  • Since QEMU in SPICE mode acts similarly to a remote desktop server, it may be more convenient to run QEMU in daemon mode with the -daemonize parameter.
  • Using Unix sockets instead of TCP ports does not involve using network stack on the host system, so it is reportedly better for performance. Example:
$ qemu-system-x86_64 -vga qxl -device virtio-serial-pci -device virtserialport,chardev=spicechannel0,name=com.redhat.spice.0 -chardev spicevmc,id=spicechannel0,name=vdagent -spice unix,addr=/tmp/vm_spice.socket,disable-ticketing

Then connect with $ remote-viewer spice+unix:///tmp/vm_spice.socket or with $ spicy --uri="spice+unix:///tmp/vm_spice.socket".

Connect to the guest with a SPICE client

A SPICE client is necessary to connect to the guest. In Arch, the following clients are available:

virt-viewer — is the recommended SPICE client by the protocol developers

run it with $ remote-viewer spice://127.0.0.1:5930 || virt-viewer

spice-gtk — is a GTK+ client which can also be used

run it with $ spicy -h 127.0.0.1 -p 5930 || spice-gtk
Tip: To connect to the guest through SSH tunelling, the following type of command can be used:
$ ssh -fL 5999:localhost:5930 my.domain.org sleep 10; spicy -h 127.0.0.1 -p 5999

This example connects spicy to the local port 5999 which is forwarded through SSH to the guest's SPICE server located at the address my.domain.org, port 5930. Note the -f option that requests ssh to execute the command sleep 10 in the background. This way, the ssh session runs while the client is active and auto-closes once the client ends.

For clients that run on smartphone or on other platforms, refer to the Other clients section in spice-space download.

SPICE support on the guest

For Arch Linux guests, for improved support for multiple monitors or clipboard sharing, the following packages should be installed:

  • spice-vdagent: Spice agent xorg client that enables copy and paste between client and X-session and more. Enable spice-vdagentd.service after installation.
  • xf86-video-qxl: Xorg X11 qxl video driver

For guests under other operating systems, refer to the Guest section in spice-space download.

Password authentication with SPICE

If you want to enable password authentication with SPICE you need to remove disable-ticketing from the -spice argument and instead add password=yourpassword. For example:

$ qemu-system-x86_64 -vga qxl -spice port=5900,password=yourpassword -device virtio-serial-pci -device virtserialport,chardev=spicechannel0,name=com.redhat.spice.0 -chardev spicevmc,id=spicechannel0,name=vdagent

Your SPICE client should now ask for the password to be able to connect to the SPICE server.

TLS encrypted communication with SPICE

You can also configure TLS encryption for communicating with the SPICE server. First, you need to have a directory which contains the following files (the names must be exactly as indicated):

  • ca-cert.pem: the CA master certificate.
  • server-cert.pem: the server certificate signed with ca-cert.pem.
  • server-key.pem: the server private key.

An example of generation of self-signed certificates with your own generated CA for your server is shown in the Spice User Manual.

Afterwards, you can run QEMU with SPICE as explained above but using the following -spice argument: -spice tls-port=5901,password=yourpassword,x509-dir=/path/to/pki_certs, where /path/to/pki_certs is the directory path that contains the three needed files shown earlier.

It is now possible to connect to the server using virt-viewer:

$ remote-viewer spice://hostname?tls-port=5901 --spice-ca-file=/path/to/ca-cert.pem --spice-host-subject="C=XX,L=city,O=organization,CN=hostname" --spice-secure-channels=all

Keep in mind that the --spice-host-subject parameter needs to be set according to your server-cert.pem subject. You also need to copy ca-cert.pem to every client to verify the server certificate.

Tip: You can get the subject line of the server certificate in the correct format for --spice-host-subject (with entries separated by commas) using the following command:
$ openssl x509 -noout -subject -in server-cert.pem | cut -d' ' -f2- | sed 's/\///' | sed 's/\//,/g'

The equivalent spice-gtk command is:

$ spicy -h hostname -s 5901 --spice-ca-file=ca-cert.pem --spice-host-subject="C=XX,L=city,O=organization,CN=hostname" --spice-secure-channels=all

VNC

One can add the -vnc :X option to have QEMU redirect the VGA display to the VNC session. Substitute X for the number of the display (0 will then listen on 5900, 1 on 5901...).

$ qemu-system-x86_64 -vnc :0

An example is also provided in the #Starting QEMU virtual machines on boot section.

Warning: The default VNC server setup does not use any form of authentication. Any user can connect from any host.

Basic password authentication

An access password can be setup easily by using the password option. The password must be indicated in the QEMU monitor and connection is only possible once the password is provided.

$ qemu-system-x86_64 -vnc :0,password -monitor stdio

In the QEMU monitor, password is set using the command change vnc password and then indicating the password.

The following command line directly runs vnc with a password:

$ printf "change vnc password\n%s\n" MYPASSWORD | qemu-system-x86_64 -vnc :0,password -monitor stdio
Note: The password is limited to 8 characters and can be guessed through brute force attack. More elaborated protection is strongly recommended for public network.

Audio

Host

The audio driver used by QEMU is set with the QEMU_AUDIO_DRV environment variable:

$ export QEMU_AUDIO_DRV=pa

Run the following command to get QEMU's configuration options related to PulseAudio:

$ qemu-system-x86_64 -audio-help | awk '/Name: pa/' RS=

The listed options can be exported as environment variables, for example:

$ export QEMU_PA_SINK=alsa_output.pci-0000_04_01.0.analog-stereo.monitor
$ export QEMU_PA_SOURCE=input

Guest

To get list of the supported emulation audio drivers:

$ qemu-system-x86_64 -soundhw help

To use e.g. hda driver for the guest use the -soundhw hda command with QEMU.

Note: Video graphic card emulated drivers for the guest machine may also cause a problem with the sound quality. Test one by one to make it work. You can list possible options with qemu-system-x86_64 -h | grep vga.

Installing virtio drivers

QEMU offers guests the ability to use paravirtualized block and network devices using the virtio drivers, which provide better performance and lower overhead.

  • A virtio block device requires the option -drive for passing a disk image, with parameter if=virtio:
$ qemu-system-x86_64 -boot order=c -drive file=disk_image,if=virtio
  • Almost the same goes for the network:
$ qemu-system-x86_64 -net nic,model=virtio
Note: This will only work if the guest machine has drivers for virtio devices. Linux does, and the required drivers are included in Arch Linux, but there is no guarantee that virtio devices will work with other operating systems.

Preparing an (Arch) Linux guest

To use virtio devices after an Arch Linux guest has been installed, the following modules must be loaded in the guest: virtio, virtio_pci, virtio_blk, virtio_net, and virtio_ring. For 32-bit guests, the specific "virtio" module is not necessary.

If you want to boot from a virtio disk, the initial ramdisk must contain the necessary modules. By default, this is handled by mkinitcpio's autodetect hook. Otherwise use the MODULES array in /etc/mkinitcpio.conf to include the necessary modules and rebuild the initial ramdisk.

/etc/mkinitcpio.conf
MODULES=(virtio virtio_blk virtio_pci virtio_net)

Virtio disks are recognized with the prefix v (e.g. vda, vdb, etc.); therefore, changes must be made in at least /etc/fstab and /boot/grub/grub.cfg when booting from a virtio disk.

Tip: When referencing disks by UUID in both /etc/fstab and bootloader, nothing has to be done.

Further information on paravirtualization with KVM can be found here.

You might also want to install qemu-guest-agent to implement support for QMP commands that will enhance the hypervisor management capabilities. After installing the package you can enable and start the qemu-ga.service.

Preparing a Windows guest

Note: The only (reliable) way to upgrade a Windows 8.1 guest to Windows 10 seems to be to temporarily choose cpu core2duo,nx for the install [1]. After the install, you may revert to other cpu settings (8/8/2015).

Block device drivers

New Install of Windows

Windows does not come with the virtio drivers. Therefore, you will need to load them during installation. There are basically two ways to do this: via Floppy Disk or via ISO files. Both images can be downloaded from the Fedora repository.

The floppy disk option is difficult because you will need to press F6 (Shift-F6 on newer Windows) at the very beginning of powering on the QEMU. This is difficult since you need time to connect your VNC console window. You can attempt to add a delay to the boot sequence. See qemu(1) for more details about applying a delay at boot.

The ISO option to load drivers is the preferred way, but it is available only on Windows Vista and Windows Server 2008 and later. The procedure is to load the image with virtio drivers in an additional cdrom device along with the primary disk device and Windows installer:

$ qemu-system-x86_64 ... \
-drive file=/path/to/primary/disk.img,index=0,media=disk,if=virtio \
-drive file=/path/to/installer.iso,index=2,media=cdrom \
-drive file=/path/to/virtio.iso,index=3,media=cdrom \
...

During the installation, the Windows installer will ask you for your Product key and perform some additional checks. When it gets to the "Where do you want to install Windows?" screen, it will give a warning that no disks are found. Follow the example instructions below (based on Windows Server 2012 R2 with Update).

  • Select the option Load Drivers.
  • Uncheck the box for "Hide drivers that aren't compatible with this computer's hardware".
  • Click the Browse button and open the CDROM for the virtio iso, usually named "virtio-win-XX".
  • Now browse to E:\viostor\[your-os]\amd64, select it, and press OK.
  • Click Next

You should now see your virtio disk(s) listed here, ready to be selected, formatted and installed to.

Change Existing Windows VM to use virtio

Modifying an existing Windows guest for booting from virtio disk is a bit tricky.

You can download the virtio disk driver from the Fedora repository.

Now you need to create a new disk image, which will force Windows to search for the driver. For example:

$ qemu-img create -f qcow2 fake.qcow2 1G

Run the original Windows guest (with the boot disk still in IDE mode) with the fake disk (in virtio mode) and a CD-ROM with the driver.

$ qemu-system-x86_64 -m 512 -drive file=windows_disk_image,if=ide -drive file=fake.qcow2,if=virtio -cdrom virtio-win-0.1-81.iso

Windows will detect the fake disk and try to find a driver for it. If it fails, go to the Device Manager, locate the SCSI drive with an exclamation mark icon (should be open), click Update driver and select the virtual CD-ROM. Do not navigate to the driver folder within the CD-ROM, simply select the CD-ROM drive and Windows will find the appropriate driver automatically (tested for Windows 7 SP1). Do not forget to select the checkbox which says to search for directories recursively.

When the installation is successful, you can turn off the virtual machine and launch it again, now with the boot disk attached in virtio mode:

$ qemu-system-x86_64 -m 512 -drive file=windows_disk_image,if=virtio
Note: If you encounter the Blue Screen of Death, make sure you did not forget the -m parameter, and that you do not boot with virtio instead of ide for the system drive before drivers are installed.

Network drivers

Installing virtio network drivers is a bit easier, simply add the -net argument as explained above.

$ qemu-system-x86_64 -m 512 -drive file=windows_disk_image,if=virtio -net nic,model=virtio -cdrom virtio-win-0.1-74.iso

Windows will detect the network adapter and try to find a driver for it. If it fails, go to the Device Manager, locate the network adapter with an exclamation mark icon (should be open), click Update driver and select the virtual CD-ROM. Do not forget to select the checkbox which says to search for directories recursively.

Balloon driver

If you want to track you guest memory state (for example via virsh command dommemstat) or change guest's memory size in runtime (you still won't be able to change memory size, but can limit memory usage via inflating balloon driver) you will need to install guest balloon driver.

For this you will need to go to Device Manager, locate PCI standard RAM Controller in System devices (or unrecognized PCI controller from Other devices) and choose Update driver. In opened window you will need to choose Browse my computer... and select the CD-ROM (and don't forget the Include subdirectories checkbox). Reboot after installation. This will install the driver and you will be able to inflate the balloon (for example via hmp command balloon memory_size, which will cause balloon to take as much memory as possible in order to shrink the guest's available memory size to memory_size). However, you still won't be able to track guest memory state. In order to do this you will need to install Balloon service properly. For that open command line as administrator, go to the CD-ROM, Balloon directory and deeper, depending on your system and architecture. Once you are in amd64 (x86) directory, run blnsrv.exe -i which will do the installation. After that virsh command dommemstat should be outputting all supported values.

Preparing a FreeBSD guest

Install the emulators/virtio-kmod port if you are using FreeBSD 8.3 or later up until 10.0-CURRENT where they are included into the kernel. After installation, add the following to your /boot/loader.conf file:

virtio_load="YES"
virtio_pci_load="YES"
virtio_blk_load="YES"
if_vtnet_load="YES"
virtio_balloon_load="YES"

Then modify your /etc/fstab by doing the following:

sed -ibak "s/ada/vtbd/g" /etc/fstab

And verify that /etc/fstab is consistent. If anything goes wrong, just boot into a rescue CD and copy /etc/fstab.bak back to /etc/fstab.

QEMU monitor

While QEMU is running, a monitor console is provided in order to provide several ways to interact with the virtual machine running. The QEMU monitor offers interesting capabilities such as obtaining information about the current virtual machine, hotplugging devices, creating snapshots of the current state of the virtual machine, etc. To see the list of all commands, run help or ? in the QEMU monitor console or review the relevant section of the official QEMU documentation.

Accessing the monitor console

When using the std default graphics option, one can access the QEMU monitor by pressing Ctrl+Alt+2 or by clicking View > compatmonitor0 in the QEMU window. To return to the virtual machine graphical view either press Ctrl+Alt+1 or click View > VGA.

However, the standard method of accessing the monitor is not always convenient and does not work in all graphic outputs QEMU supports. Alternative options of accessing the monitor are described below:

  • telnet: Run QEMU with the -monitor telnet:127.0.0.1:port,server,nowait parameter. When the virtual machine is started you will be able to access the monitor via telnet:
 $ telnet 127.0.0.1 port
Note: If 127.0.0.1 is specified as the IP to listen it will be only possible to connect to the monitor from the same host QEMU is running on. If connecting from remote hosts is desired, QEMU must be told to listen 0.0.0.0 as follows: -monitor telnet:0.0.0.0:port,server,nowait. Keep in mind that it is recommended to have a firewall configured in this case or make sure your local network is completely trustworthy since this connection is completely unauthenticated and unencrypted.
  • UNIX socket: Run QEMU with the -monitor unix:socketfile,server,nowait parameter. Then you can connect with either socat or openbsd-netcat.

For example, if QEMU is run via:

 $ qemu-system-x86_64 [...] -monitor unix:/tmp/monitor.sock,server,nowait [...]

It is possible to connect to the monitor with:

 $ socat - UNIX-CONNECT:/tmp/monitor.sock

Or with:

 $ nc -U /tmp/monitor.sock
  • TCP: You can expose the monitor over TCP with the argument -monitor tcp:127.0.0.1:port,server,nowait. Then connect with netcat, either openbsd-netcat or gnu-netcat by running:
 $ nc 127.0.0.1 port
Note: In order to be able to connect to the tcp socket from other devices other than the same host QEMU is being run on you need to listen to 0.0.0.0 like explained in the telnet case. The same security warnings apply in this case as well.
  • Standard I/O: It is possible to access the monitor automatically from the same terminal QEMU is being run by running it with the argument -monitor stdio.

Sending keyboard presses to the virtual machine using the monitor console

Some combinations of keys may be difficult to perform on virtual machines due to the host intercepting them instead in some configurations (a notable example is the Ctrl+Alt+F* key combinations, which change the active tty). To avoid this problem, the problematic combination of keys may be sent via the monitor console instead. Switch to the monitor and use the sendkey command to forward the necessary keypresses to the virtual machine. For example:

 (qemu) sendkey ctrl-alt-f2

Creating and managing snapshots via the monitor console

Note: This feature will only work when the virtual machine disk image is in qcow2 format. It will not work with raw images.

It is sometimes desirable to save the current state of a virtual machine and having the possibility of reverting the state of the virtual machine to that of a previously saved snapshot at any time. The QEMU monitor console provides the user with the necessary utilities to create snapshots, manage them, and revert the machine state to a saved snapshot.

  • Use savevm name in order to create a snapshot with the tag name.
  • Use loadvm name to revert the virtual machine to the state of the snapshot name.
  • Use delvm name to delete the snapshot tagged as name.
  • Use info snapshots to see a list of saved snapshots. Snapshots are identified by both an auto-incremented ID number and a text tag (set by the user on snapshot creation).

Running the virtual machine in immutable mode

It is possible to run a virtual machine in a frozen state so that all changes will be discarded when the virtual machine is powered off just by running QEMU with the -snapshot parameter. When the disk image is written by the guest, changes will be saved in a temporary file in /tmp and will be discarded when QEMU halts.

However, if a machine is running in frozen mode it is still possible to save the changes to the disk image if it is afterwards desired by using the monitor console and running the following command:

 (qemu) commit all

If snapshots are created when running in frozen mode they will be discarded as soon as QEMU is exited unless changes are explicitly commited to disk, as well.

Pause and power options via the monitor console

Some operations of a physical machine can be emulated by QEMU using some monitor commands:

  • system_powerdown will send an ACPI shutdown request to the virtual machine. This effect is similar to the power button in a physical machine.
  • system_reset will reset the virtual machine similarly to a reset button in a physical machine. This operation can cause data loss and file system corruption since the virtual machine is not cleanly restarted.
  • stop will pause the virtual machine.
  • cont will resume a virtual machine previously paused.

Taking screenshots of the virtual machine

Screenshots of the virtual machine graphic display can be obtained in the PPM format by running the following command in the monitor console:

 (qemu) screendump file.ppm

QEMU machine protocol

The QEMU machine protocol (QMP) is a JSON-based protocol which allows applications to control a QEMU instance. Similarly to the #QEMU monitor it offers ways to interact with a running machine and the JSON protocol allows to do it programmatically. The description of all the QMP commands can be found in qmp-commands.

Start QMP

The usual way to control the guest using the QMP protocol, is to open a TCP socket when launching the machine using the -qmp option. Here it is using for example the TCP port 4444:

 $ qemu-system-x86_64 [...] -qmp tcp:localhost:4444,server,nowait

Then one way to communicate with the QMP agent is to use netcat:

nc localhost 4444
{"QMP": {"version": {"qemu": {"micro": 0, "minor": 1, "major": 3}, "package": ""}, "capabilities": []} } 

At this stage, the only command that can be recognized is qmp_capabilities, so that QMP enters into command mode. Type:

 {"execute": "qmp_capabilities"}

Now, QMP is ready to receive commands, to retrieve the list of recognized commands, use:

 {"execute": "query-commands"}

Live merging of child image into parent image

It is possible to merge a running snapshot into its parent by issuing a block-commit command. In its simplest form the following line will commit the child into its parent:

 {"execute": "block-commit", "arguments": {"device": "devicename"}}

Upon reception of this command, the handler looks for the base image and converts it from read only to read write mode and then runs the commit job.

Once the block-commit operation has completed, the event BLOCK_JOB_READY will be emitted, signalling that the synchronization has finished. The job can then be gracefully completed by issuing the command block-job-complete:

 {"execute": "block-job-complete", "arguments": {"device": "devicename"}}

Until such a command is issued, the commit operation remains active. After successful completion, the base image remains in read write mode and becomes the new active layer. On the other hand, the child image becomes invalid and it is the responsibility of the user to clean it up.

Tip: The list of device and their names can be retrieved by executing the command query-block and parsing the results. The device name is in the device field, for example ide0-hd0 for the hard disk in this example:
{"execute": "query-block"}
{"return": [{"io-status": "ok", "device": "ide0-hd0", "locked": false, "removable": false, "inserted": {"iops_rd": 0, "detect_zeroes": "off", "image": {"backing-image": {"virtual-size": 27074281472, "filename": "parent.qcow2", ... } 

Live creation of a new snapshot

To create a new snapshot out of a running image, run the command:

 {"execute": "blockdev-snapshot-sync", "arguments": {"device": "devicename","snapshot-file": "new_snapshot_name.qcow2"}}

This creates an overlay file named new_snapshot_name.qcow2 which then becomes the new active layer.

Tips and tricks

Improve virtual machine performance

There are a number of techniques that you can use to improve the performance of the virtual machine. For example:

  • Use the -cpu host option to make QEMU emulate the host's exact CPU. If you do not do this, it may be trying to emulate a more generic CPU.
  • Especially for Windows guests, enable Hyper-V enlightenments: -cpu host,hv_relaxed,hv_spinlocks=0x1fff,hv_vapic,hv_time.
  • If the host machine has multiple cores, assign the guest more cores using the -smp option.
  • Make sure you have assigned the virtual machine enough memory. By default, QEMU only assigns 128 MiB of memory to each virtual machine. Use the -m option to assign more memory. For example, -m 1024 runs a virtual machine with 1024 MiB of memory.
  • Apply #Enabling KVM: add -enable-kvm to the QEMU start command you use.
  • If supported by drivers in the guest operating system, use virtio for network and/or block devices. For example:
$ qemu-system-x86_64 -net nic,model=virtio -net tap,if=tap0,script=no -drive file=disk_image,media=disk,if=virtio
  • Use TAP devices instead of user-mode networking. See #Tap networking with QEMU.
  • If the guest OS is doing heavy writing to its disk, you may benefit from certain mount options on the host's file system. For example, you can mount an ext4 file system with the option barrier=0. You should read the documentation for any options that you change because sometimes performance-enhancing options for file systems come at the cost of data integrity.
  • If you have a raw disk image, you may want to disable the cache:
$ qemu-system-x86_64 -drive file=disk_image,if=virtio,cache=none
  • Use the native Linux AIO:
$ qemu-system-x86_64 -drive file=disk_image,if=virtio,aio=native,cache.direct=on
  • If you are running multiple virtual machines concurrently that all have the same operating system installed, you can save memory by enabling kernel same-page merging. See #Enabling KSM.
  • In some cases, memory can be reclaimed from running virtual machines by running a memory ballooning driver in the guest operating system and launching QEMU using -device virtio-balloon.
  • It is possible to use a emulation layer for an ICH-9 AHCI controller (although it may be unstable). The AHCI emulation supports NCQ, so multiple read or write requests can be outstanding at the same time:
$ qemu-system-x86_64 -drive id=disk,file=disk_image,if=none -device ich9-ahci,id=ahci -device ide-drive,drive=disk,bus=ahci.0

See http://www.linux-kvm.org/page/Tuning_KVM for more information.

Starting QEMU virtual machines on boot

With libvirt

If a virtual machine is set up with libvirt, it can be configured with virsh autostart or through the virt-manager GUI to start at host boot by going to the Boot Options for the virtual machine and selecting "Start virtual machine on host boot up".

With systemd service

To run QEMU VMs on boot, you can use following systemd unit and config.

/etc/systemd/system/qemu@.service
[Unit]
Description=QEMU virtual machine

[Service]
Environment="type=system-x86_64" "haltcmd=kill -INT $MAINPID"
EnvironmentFile=/etc/conf.d/qemu.d/%i
ExecStart=/usr/bin/qemu-${type} -name %i -nographic $args
ExecStop=/bin/sh -c ${haltcmd}
TimeoutStopSec=30
KillMode=none

[Install]
WantedBy=multi-user.target
Note: According to systemd.service(5) and 5 man pages it is necessary to use the KillMode=none option. Otherwise the main qemu process will be killed immediately after the ExecStop command quits (it simply echoes one string) and your quest system will not be able to shutdown correctly.

Then create per-VM configuration files, named /etc/conf.d/qemu.d/vm_name, with the variables type, args and altcmd set. Example configs:

/etc/conf.d/qemu.d/one
type="system-x86_64"

args="-enable-kvm -m 512 -hda /dev/vg0/vm1 -net nic,macaddr=DE:AD:BE:EF:E0:00 \
 -net tap,ifname=tap0 -serial telnet:localhost:7000,server,nowait,nodelay \
 -monitor telnet:localhost:7100,server,nowait,nodelay -vnc :0"

haltcmd="echo 'system_powerdown' | nc localhost 7100" # or netcat/ncat

# You can use other ways to shut down your VM correctly
#haltcmd="ssh powermanager@vm1 sudo poweroff"
/etc/conf.d/qemu.d/two
args="-enable-kvm -m 512 -hda /srv/kvm/vm2.img -net nic,macaddr=DE:AD:BE:EF:E0:01 \
 -net tap,ifname=tap1 -serial telnet:localhost:7001,server,nowait,nodelay \
 -monitor telnet:localhost:7101,server,nowait,nodelay -vnc :1"

haltcmd="echo 'system_powerdown' | nc localhost 7101"

The description of the variables is the following:

  • type - QEMU binary to call. If specified, will be prepended with /usr/bin/qemu- and that binary will be used to start the VM.
  • args - QEMU command line to start with. Will always be prepended with -name ${vm} -nographic.
  • haltcmd - Command to shut down a VM safely. In this example, the QEMU monitor is exposed via telnet using -monitor telnet:.. and the VMs are powered off via ACPI by sending system_powerdown to monitor with the nc command. You can use SSH or some other ways as well.

To set which virtual machines will start on boot-up, enable the qemu@vm_name.service systemd unit.

Mouse integration

To prevent the mouse from being grabbed when clicking on the guest operating system's window, add the options -usb -device usb-tablet. This means QEMU is able to report the mouse position without having to grab the mouse. This also overrides PS/2 mouse emulation when activated. For example:

$ qemu-system-x86_64 -hda disk_image -m 512 -usb -device usb-tablet

If that does not work, try using -vga qxl parameter, also look at the instructions #Mouse cursor is jittery or erratic.

Pass-through host USB device

It is possible to access the physical device connected to a USB port of the host from the guest. The first step is to identify where the device is connected, this can be found running the lsusb command. For example:

lsusb
...
Bus 003 Device 007: ID 0781:5406 SanDisk Corp. Cruzer Micro U3

The outputs in bold above will be useful to identify respectively the host_bus and host_addr or the vendor_id and product_id.

In qemu, the idea is to emulate an EHCI (USB 2) or XHCI (USB 3) controller with the option -device usb-ehci,id=ehci or -device qemu-xhci,id=xhci respectively and then attach the physical device to it with the option -device usb-host,... We will consider that controller_id is either ehci or xhci for the rest of this section.

Then, there are two ways to connect to the USB of the host with qemu:

  1. Identify the device and connect to it on any bus and address it is attached to on the host, the generic syntax is:
    -device usb-host,bus=controller_id.0,vendorid=0xvendor_id,productid=0xproduct_id
    Applied to the device used in the example above, it becomes:
    -device usb-ehci,id=ehci -device usb-host,bus=ehci.0,vendorid=0x0781,productid=0x5406
    One can also add the ...,port=port_number setting to the previous option to specify in which physical port of the virtual controller the device should be attached, useful in the case one want to add multiple usb devices to the VM.
  2. Attach whatever is connected to a given USB bus and address, the syntax is:
    -device usb-host,bus=controller_id.0,hostbus=host_bus,host_addr=host_addr
    Applied to the bus and the address in the example above, it becomes:
    -device usb-ehci,id=ehci -device usb-host,bus=ehci.0,hostbus=3,hostaddr=7
Note: If you encounter permission errors when running QEMU, see udev#About udev rules for information on how to set permissions of the device.

USB redirection with SPICE

When using #SPICE it is possible to redirect USB devices from the client to the virtual machine without needing to specify them in the QEMU command. It is possible to configure the number of USB slots available for redirected devices (the number of slots will determine the maximum number of devices which can be redirected simultaneously). The main advantages of using SPICE for redirection compared to the previously-mentioned -usbdevice method is the possibility of hot-swapping USB devices after the virtual machine has started, without needing to halt it in order to remove USB devices from the redirection or adding new ones. This method of USB redirection also allows us to redirect USB devices over the network, from the client to the server. In summary, it is the most flexible method of using USB devices in a QEMU virtual machine.

We need to add one EHCI/UHCI controller per available USB redirection slot desired as well as one SPICE redirection channel per slot. For example, adding the following arguments to the QEMU command you use for starting the virtual machine in SPICE mode will start the virtual machine with three available USB slots for redirection:

-device ich9-usb-ehci1,id=usb \
-device ich9-usb-uhci1,masterbus=usb.0,firstport=0,multifunction=on \
-device ich9-usb-uhci2,masterbus=usb.0,firstport=2 \
-device ich9-usb-uhci3,masterbus=usb.0,firstport=4 \
-chardev spicevmc,name=usbredir,id=usbredirchardev1 -device usb-redir,chardev=usbredirchardev1,id=usbredirdev1 \
-chardev spicevmc,name=usbredir,id=usbredirchardev2 -device usb-redir,chardev=usbredirchardev2,id=usbredirdev2 \
-chardev spicevmc,name=usbredir,id=usbredirchardev3 -device usb-redir,chardev=usbredirchardev3,id=usbredirdev3

See SPICE/usbredir for more information.

Both spicy from spice-gtk (Input > Select USB Devices for redirection) and remote-viewer from virt-viewer (File > USB device selection) support this feature. Please make sure that you have installed the necessary SPICE Guest Tools on the virtual machine for this functionality to work as expected (see the #SPICE section for more information).

Warning: Keep in mind that when a USB device is redirected from the client, it will not be usable from the client operating system itself until the redirection is stopped. It is specially important to never redirect the input devices (namely mouse and keyboard), since it will be then difficult to access the SPICE client menus to revert the situation, because the client will not respond to the input devices after being redirected to the virtual machine.

Enabling KSM

Kernel Samepage Merging (KSM) is a feature of the Linux kernel that allows for an application to register with the kernel to have its pages merged with other processes that also register to have their pages merged. The KSM mechanism allows for guest virtual machines to share pages with each other. In an environment where many of the guest operating systems are similar, this can result in significant memory savings.

Note: Although KSM may reduce memory usage, it may increase CPU usage. Also note some security issues may occur, see Wikipedia:Kernel same-page merging.

To enable KSM:

# echo 1 > /sys/kernel/mm/ksm/run

To make it permanent, use systemd's temporary files:

/etc/tmpfiles.d/ksm.conf
w /sys/kernel/mm/ksm/run - - - - 1

If KSM is running, and there are pages to be merged (i.e. at least two similar VMs are running), then /sys/kernel/mm/ksm/pages_shared should be non-zero. See https://www.kernel.org/doc/Documentation/vm/ksm.txt for more information.

Tip: An easy way to see how well KSM is performing is to simply print the contents of all the files in that directory:
$ grep . /sys/kernel/mm/ksm/*

Multi-monitor support

The Linux QXL driver supports four heads (virtual screens) by default. This can be changed via the qxl.heads=N kernel parameter.

The default VGA memory size for QXL devices is 16M (VRAM size is 64M). This is not sufficient if you would like to enable two 1920x1200 monitors since that requires 2 × 1920 × 4 (color depth) × 1200 = 17.6 MiB VGA memory. This can be changed by replacing -vga qxl by -vga none -device qxl-vga,vgamem_mb=32. If you ever increase vgamem_mb beyond 64M, then you also have to increase the vram_size_mb option.

Copy and paste

One way to share the clipboard between the host and the guest is to enable the SPICE remote desktop protocol and access the client with a SPICE client. One needs to follow the steps described in #SPICE. A guest run this way will support copy paste with the host.

Windows-specific notes

QEMU can run any version of Windows from Windows 95 through Windows 10.

It is possible to run Windows PE in QEMU.

Fast startup

Note: An administrator account is required to change power settings.

For Windows 8 (or later) guests it is better to disable "Turn on fast startup (recommended)" from the Power Options of the Control Panel as explained in the following forum page, as it causes the guest to hang during every other boot.

Fast Startup may also need to be disabled for changes to the -smp option to be properly applied.

Remote Desktop Protocol

If you use a MS Windows guest, you might want to use RDP to connect to your guest VM. If you are using a VLAN or are not in the same network as the guest, use:

$ qemu-system-x86_64 -nographic -net user,hostfwd=tcp::5555-:3389

Then connect with either rdesktop or freerdp to the guest. For example:

$ xfreerdp -g 2048x1152 localhost:5555 -z -x lan

Clone Linux system installed on physical equipment

Linux system installed on physical equipment can be cloned for running on QEMU vm. See Clone Linux system from hardware for QEMU virtual machine

Troubleshooting

Mouse cursor is jittery or erratic

If the cursor jumps around the screen uncontrollably, entering this on the terminal before starting QEMU might help:

$ export SDL_VIDEO_X11_DGAMOUSE=0

If this helps, you can add this to your ~/.bashrc file.

No visible Cursor

Add -show-cursor to QEMU's options to see a mouse cursor.

If that still does not work, make sure you have set your display device appropriately, for example: -vga qxl.

Two different mouse cursors are visible

Apply the tip #Mouse integration.

Keyboard issues when using VNC

When using VNC, you might experience keyboard problems described (in gory details) here. The solution is not to use the -k option on QEMU, and to use gvncviewer from gtk-vnc. See also this message posted on libvirt's mailing list.

Keyboard seems broken or the arrow keys do not work

Should you find that some of your keys do not work or "press" the wrong key (in particular, the arrow keys), you likely need to specify your keyboard layout as an option. The keyboard layouts can be found in /usr/share/qemu/keymaps.

$ qemu-system-x86_64 -k keymap disk_image

Guest display stretches on window resize

To restore default window size, press Ctrl+Alt+u.

ioctl(KVM_CREATE_VM) failed: 16 Device or resource busy

If an error message like this is printed when starting QEMU with -enable-kvm option:

ioctl(KVM_CREATE_VM) failed: 16 Device or resource busy
failed to initialize KVM: Device or resource busy

that means another hypervisor is currently running. It is not recommended or possible to run several hypervisors in parallel.

libgfapi error message

The error message displayed at startup:

Failed to open module: libgfapi.so.0: cannot open shared object file: No such file or directory

Install glusterfs or ignore the error message as GlusterFS is a optional dependency.

Kernel panic on LIVE-environments

If you start a live-environment (or better: booting a system) you may encounter this:

[ end Kernel panic - not syncing: VFS: Unable to mount root fs on unknown block(0,0)

or some other boot hindering process (e.g. cannot unpack initramfs, cant start service foo). Try starting the VM with the -m VALUE switch and an appropriate amount of RAM, if the ram is to low you will probably encounter similar issues as above/without the memory-switch.

Windows 7 guest suffers low-quality sound

Using the hda audio driver for Windows 7 guest may result in low-quality sound. Changing the audio driver to ac97 by passing the -soundhw ac97 arguments to QEMU and installing the AC97 driver from Realtek AC'97 Audio Codecs in the guest may solve the problem. See Red Hat Bugzilla – Bug 1176761 for more information.

Could not access KVM kernel module: Permission denied

If you encounter the following error:

 libvirtError: internal error: process exited while connecting to monitor: Could not access KVM kernel module: Permission denied failed to initialize KVM: Permission denied

Systemd 234 assign it a dynamic id to group kvm (see bug). A workground for avoid this error, you need edit the file /etc/libvirt/qemu.conf and change the line:

 group = "78"

to

 group = "kvm"

"System Thread Exception Not Handled" when booting a Windows VM

Windows 8 or Windows 10 guests may raise a generic compatibility exception at boot, namely "System Thread Exception Not Handled", which tends to be caused by legacy drivers acting strangely on real machines. On KVM machines this issue can generally be solved by setting the CPU model to core2duo.

Certain Windows games/applications crashing/causing a bluescreen

Occasionally, applications running in the VM may crash unexpectedly, whereas they would run normally on a physical machine. If, while running dmesg -wH, you encounter an error mentioning MSR, the reason for those crashes is that KVM injects a General protection fault (GPF) when the guest tries to access unsupported Model-specific registers (MSRs) - this often results in guest applications/OS crashing. A number of those issues can be solved by passing the ignore_msrs=1 option to the KVM module, which will ignore unimplemented MSRs.

/etc/modprobe.d/kvm.conf
...
options kvm ignore_msrs=1
...

Cases where adding this option might help:

  • GeForce Experience complaining about an unsupported CPU being present.
  • StarCraft 2 and L.A. Noire reliably blue-screening Windows 10 with KMODE_EXCEPTION_NOT_HANDLED. The blue screen information does not identify a driver file in these cases.
Warning: While this is normally safe and some applications might not work without this, silently ignoring unknown MSR accesses could potentially break other software within the VM or other VMs.

Applications in the VM experience long delays or take a long time to start

This may be caused by insufficient available entropy in the VM. Consider allowing the guest to access the hosts's entropy pool by adding a VirtIO RNG device to the VM, or by installing an entropy generating daemon such as Haveged.

Anecdotally, OpenSSH takes a while to start accepting connections under insufficient entropy, without the logs revealing why.

High interrupt latency and microstuttering

This problem manifests itself as small pauses (stutters) and is particularly noticeable in graphics-intensive applications, such as games. One of the causes is CPU power saving features, which are controlled by CPU frequency scaling. Change this to performance for all processor cores.

See also