Enabling Zoned Block Device Support

Kernel Configuration

Several kernel compilation configuration options control zoned block device support features.

Block Layer

Zoned Block Devices Core Support

To allow exposing supported zoned block devices as block device files, the block layer configuration option CONFIG_BLK_DEV_ZONED must be enabled. This option is part of the Enable the block layer top menu of make menuconfig.

config-zbd
Block layer zoned block device support option with make menuconfig

Without this configuration option set, users will not have access to the ZBD interface and support for zoned block devices will be disabled in all kernel subsystems (I/O schedulers, device mapper and file systems) that include support for these devices.

Write Ordering Control

Write ordering control is achieved through the deadline (legacy single queue block I/O path) and mq-deadline (multi-queue block I/O path) block I/O scheduler (see Write Ordering Control). deadline and mq-deadline zoned block device support is automatically enabled if the CONFIG_BLK_DEV_ZONED configuration option is set.

Enabling this scheduler is mandatory for zoned block devices. This is controlled with the CONFIG_MQ_IOSCHED_DEADLINE option for mq-deadline and with the CONFIG_IOSCHED_DEADLINE option for deadline. Either option can be selected from the IO Schedulers top menu.

config-sched
I/O scheduler configuration with make menuconfig

With the introduction of kernel version 5.0 and the removal of the block layer legacy single queue I/O path, only the mq-deadline scheduler remains. Since kernel version 5.2, the selection of the CONFIG_MQ_IOSCHED_DEADLINE option is automatic when the CONFIG_BLK_DEV_ZONED configuration option is set.

Device Drivers Configuration

null_blk Logical Device

Support for the zoned block device emulation with the null_blk device driver zoned mode is automatically enabled with the CONFIG_BLK_DEV_ZONED configuration option.

ZBC and ZAC Hard-Disks Support

The SCSI subsystem support for ZBC and ZAC SMR disks is automatically enabled with the CONFIG_BLK_DEV_ZONED configuration option.

NVMe Zoned Namespace Solid State Disks Support

NVM Express Zoned Namespace Command Set depends on CONFIG_BLK_DEV_ZONED and CONFIG_NVME_CORE and is automatically built if both configuration options are enabled.

The driver requires the device to support the Zone Append command to successfully bind to a zoned namespace, and does not support Zone Excursions. See Zoned Namespace (ZNS) SSDs for more details about these features.

Device Mapper

Zoned block device support for the device mapper subsystem is automatically enabled when the CONFIG_BLK_DEV_ZONED option is set. This will enables support for the dm-linear and dm-flakey targets. However, the dm-zoned device mapper target must be enabled to be usable.

Enabling the dm-zoned target can be done by selecting the CONFIG_DM_ZONED option from the menu Device Drivers → Multiple devices driver support (RAID and LVM) → Device mapper support → Drive-managed zoned block device target support.

config-dm
dm-zoned device mapper target configuration with make menuconfig

File Systems

f2fs

Support for zoned block devices in the f2fs file system is automatically enabled with the CONFIG_BLK_DEV_ZONED configuration option.

zonefs

Enabling compilation of the zonefs file system is done by selecting the CONFIG_ZONEFS_FS option from the menu File systems -> zonefs filesystem support. This option is available only and only if the CONFIG_BLK_DEV_ZONED option is set to enable zoned block device support.

config-zonefs
zonefs filesystem configuration with make menuconfig

Kernel Compilation

After completing the kernel configuration to enable zoned block device support, the kernel compilation process does not differ from building a kernel without zoned block device support. That is, the following commands will build the kernel.

$ make all

The kernel build infrastructure also enables build .rpm or i.deb packages. To build RPM packages, the following command is used.

$ make rpm-pkg

Kernel Installation

Similarly to the compilation process, installing a zoned block device enabled kernel follows the same procedure as a regular kernel. That is, for a local installation, the following command can be used.

$ sudo make modules_install install

This must be followed eventually by the system boot loader configuration as required (or not) by the distribution used.

The host system can then be restarted to execute the kernel enabling zoned block device support.

One additional step is however highly recommended: reinstalling the kernel headers. Doing so, the file /usr/include/linux/blkzoned.h will be installed, thus allowing applications to be compiled against the zoned block device API supported by the kernel.

Installing the kernel user header files is done using the following command.

$ sudo make headers_install

See the kernel make help output for more information on this directive.

With the kernel user header files installed, it is recommended to recompile from source any package that will be used to manage and access zoned block devices. In particular, recompiling and re-instlalling Linux system utilities is highly recommended as many other packages rely on util-linux zoned block device features (e.g. file systems formatting tools through libblkid).

The kernel installation and user header files installation can be simplified by using the RPM packages generated with the make rpm-pkg command. Installing all the packages generated will install the kernel core itself, the associated driver modules and the user API herder files. The RPM package kernel-headers-<version>.<arch>.rpm must be installed for the kernel user API header files to be updated.