QEMU and KVM

QEMU is a generic machine emulator and virtualizer. QEMU also provides the userspace components of the widely used KVM (Kernel-based Virtual Machine).

QEMU and zoned block devices

Host managed SMR disks can be directly attached to a QEMU guest for running applications in a virtual machine environment. This is especially useful in the case of software and kernel development and tests.

There are two supported methods to attach Host managed SMR zoned disks to a QEMU guest: virtio-scsi and vhost-scsi.

Furthermore, QEMU also allows emulating NVMe devices implementing zoned namespaces.

QEMU virtio-scsi

This is the simplest method to attach a zoned block device for access from a QEMU guest. To do so, the QEMU option virtio-scsi-pci is used after defining a virtual PCI bus and SCSI host.

For instance, the following command will run QEMU with 

# qemu-kvm (your options) \
    -device pcie-root-port,bus=pcie.0,id=rp1 \
    -device virtio-scsi-pci,bus=pcie.0,id=scsi0 \
    -drive file=/dev/sdf,format=raw,if=none,id=zbc0 \
    -device scsi-block,bus=scsi0.0,drive=zbc0

The first line -device pcie-root-port,bus=pcie.0,id=rp1 creates a PCIe root complex. The second line -device virtio-scsi-pci,bus=pcie.0,id=scsi0 defines a virtio adapter connected to the PCIe bus previously defined. Finally, the last 2 lines -drive file=/dev/sdf,format=raw,if=none,id=zbc0 and -device scsi-block,bus=scsi0.0,drive=zbc0 define the device connected to the virtio SCSI adapter. In this example, since -device scsi-block is used, the host device to attach is specified using the device block device file (/dev/sdf) in this example.

See here for a detailed description of these options.

Though this method allows attaching a zoned block disk to a QEMU guest, SCSI command sense data is not processed correctly in QEMU versions prior to version 4.1. This causes the guest operating system to hang if the guest attempts to access a command sense data (for instance upon command failures). This attachment method should thus be avoided with versions of QEMU older than version 4.1.

To avoid the sense data problem with QEMU versions preceding version 4.1, a zoned disk can be attached to a guest using the device SG node file as a specifier. In this case, the option -device scsi-generic must be used. The command line is changed as follows.

# qemu-kvm (your options) \
    -device pcie-root-port,bus=pcie.0,id=rp1 \
    -device virtio-scsi-pci,bus=pcie.0,id=scsi0 \
    -drive file=/dev/sg5,format=raw,if=none,id=zbc0 \
    -device scsi-generic,bus=scsi0.0,drive=zbc0

On the host, the correspondance between a block device file and its SG node file can be discovered using the lsscsi command.

# lsscsi -g
...
[10:0:1:0]   zbc     HGST     HSH721415AL42M0  a250  /dev/sdf   /dev/sg5
...

Once the guest operating system is started, attachment of the host device can be checked with any of the methods shown here. For instance, the output of the lsscsi command will be as follows with the above example setup.

# lsscsi -g
[0:0:0:0]    disk    ATA      QEMU HARDDISK    2.5+  /dev/sda   /dev/sg0
[6:0:1:0]    zbc     HGST     HSH721415AL42M0  a250  /dev/sdb   /dev/sg1

QEMU vhost-scsi

This attachment method uses a fabric module in the host kernel to provide KVM guests with a fast virtio-based connection to SCSI LUNs. This method cannot be used without QEMU KVM acceleration.

Enabling the host vhost target module

The host kernel configuration must have the CONFIG_VHOST_SCSI option enabled. This option is found in the top level Virtualization menu of the kernel configuration.


config-vhost-scsi
vhost-scsi support option with make menuconfig

To allow attaching physical disks as well as tcmu-runner emulated ZBC disks, the kernel configuration option COFNGI_TCM_PSCSI should also be enbaled. This option can be found in the menu Device Drivers -> Generic Target Core Mod (TCM) and ConfigFS Infrastructure.

config-vhost-scsi
pSCSI TCM support option with make menuconfig

Attaching a host physical disk

To attach to a virtual machine guest a zoned device, a virtual TCM SAS adapter must first be prepared using the targetcli tool. The device to attach must be specified using a block device file. The example below illustrates this operation for the disk /dev/sdf.

# targetcli
targetcli shell version 2.1.fb49
Copyright 2011-2013 by Datera, Inc and others.
For help on commands, type 'help'.

/> cd backstores/pscsi
/backstores/pscsi> create name=disk1 dev=/dev/sdf
Note: block backstore recommended for SCSI block devices
Created pscsi storage object disk1 using /dev/sdf
/backstores/pscsi> cd /vhost
/vhost> create
Created target naa.5001405a160fe2e1.
Created TPG 1.
/vhost/naa.5001405a160fe2e1> cd /vhost/naa.5001405a160fe2e1/tpg1/luns
/vhost/naa.50...2e1/tpg1/luns> create /backstores/pscsi/disk1
Created LUN 0.
/vhost/naa.50...2e1/tpg1/luns> cd /
/> ls
o- / ..................................................................... [...]
  o- backstores .......................................................... [...]
  | o- block .............................................. [Storage Objects: 0]
  | o- fileio ............................................. [Storage Objects: 0]
  | o- pscsi .............................................. [Storage Objects: 1]
  | | o- disk1 ............................................ [/dev/sdf activated]
  | |   o- alua ............................................... [ALUA Groups: 0]
  | o- ramdisk ............................................ [Storage Objects: 0]
  | o- user:fbo ........................................... [Storage Objects: 0]
  | o- user:rbd ........................................... [Storage Objects: 0]
  | o- user:zbc ........................................... [Storage Objects: 0]
  o- iscsi ........................................................ [Targets: 0]
  o- loopback ..................................................... [Targets: 0]
  o- vhost ........................................................ [Targets: 1]
    o- naa.5001405a160fe2e1 .......................................... [TPGs: 1]
      o- tpg1 .............................. [naa.500140565cd16730, no-gen-acls]
        o- acls ...................................................... [ACLs: 0]
        o- luns ...................................................... [LUNs: 1]
          o- lun0 .............................. [pscsi/disk1 (/dev/sdf) (None)]
/> exit

The World-Wide port name assigned by targetcli can then be used to specify the device to attach on QEMU command line.

# qemu-kvm (your options) \
    -device pcie-root-port,bus=pcie.0,id=rp1 \
    -device vhost-scsi-pci,wwpn=naa.5001405a160fe2e1,bus=pcie.0

The attached disk can then be seen from the guest OS using (for instance) the lsscsi command.

# lsscsi -g
[0:0:0:0]    disk    ATA      QEMU HARDDISK    2.5+  /dev/sda   /dev/sg0
[6:0:1:0]    zbc     HGST     HSH721415AL42M0  a250  /dev/sdb   /dev/sg1

Attaching an emulated ZBC disk

tcmu-runner can be used to create emulated ZBC host managed SCSI disks. The emulated disk created can be used either locally on the host using the loopback fabric adapter, as explained here.

Similarly to a physical device (previous section), the emulated ZBC disk can also be attached to a vhost virtual adapter for use within a KVM guest operating system. The following example illustrates this procedure, creating a small 20GB host managed SCSI disk with 256 MB zones including 10 conventional zones.

# targetcli
targetcli shell version 2.1.fb49
Copyright 2011-2013 by Datera, Inc and others.
For help on commands, type 'help'.

/> cd /backstores/user:zbc
/backstores/user:zbc> create name=zbc0 size=20G cfgstring=model-HM/zsize-256/conv-10@/var/local/zbc0.raw
Created user-backed storage object zbc0 size 21474836480.
/backstores/user:zbc> cd /vhost
/vhost> create
Created target naa.5001405a0776dce3.
Created TPG 1.
/vhost> /vhost/naa.5001405a0776dce3/tpg1/luns create /backstores/user:zbc/zbc0
Created LUN 0.
/vhost> cd /
/> ls
o- / ..................................................................... [...]
  o- backstores .......................................................... [...]
  | o- block .............................................. [Storage Objects: 0]
  | o- fileio ............................................. [Storage Objects: 0]
  | o- pscsi .............................................. [Storage Objects: 0]
  | o- ramdisk ............................................ [Storage Objects: 0]
  | o- user:fbo ........................................... [Storage Objects: 0]
  | o- user:rbd ........................................... [Storage Objects: 0]
  | o- user:zbc ........................................... [Storage Objects: 1]
  |   o- zbc0  [model-HM/zsize-256/conv-10@/var/local/zbc0.raw (20.0GiB) activated]
  |     o- alua ............................................... [ALUA Groups: 1]
  |       o- default_tg_pt_gp ................... [ALUA state: Active/optimized]
  o- iscsi ........................................................ [Targets: 0]
  o- loopback ..................................................... [Targets: 0]
  o- vhost ........................................................ [Targets: 1]
    o- naa.5001405a0776dce3 .......................................... [TPGs: 1]
      o- tpg1 .............................. [naa.500140533e375d94, no-gen-acls]
        o- acls ...................................................... [ACLs: 0]
        o- luns ...................................................... [LUNs: 1]
          o- lun0 ............................... [user/zbc0 (default_tg_pt_gp)]
/> exit

The virtual machine can then be started with the emulated ZBC disk attached using the World-Wide port name assigned by targetcli.

# qemu-kvm (your options) \
    -device pcie-root-port,bus=pcie.0,id=rp1 \
    -device vhost-scsi-pci,wwpn=naa.5001405a0776dce3,bus=pcie.0

The disk is listed on the guest with tools such as lsscsi.

# lsscsi -g
[0:0:0:0]    disk    ATA      QEMU HARDDISK    2.5+  /dev/sda   /dev/sg0
[6:0:1:0]    zbc     LIO-ORG  TCMU ZBC device  0002  /dev/sdb   /dev/sg1

QEMU NVMe ZNS Device emulation

This article describes in details, with examples, how QEMU can be configured to create an emulated NVMe ZNS namespace visible by the guest operating system.