Simple null block device manager script, useful for zoned device testing. The devices are managed by configfs files, this script wraps the basic operations for convenience. The null block device is usually memory-backed so the capacity is limited but it should be sufficient to create 1GiB or 2GiB devices. The benefit is flexible zone size that could be otherwise unavailable with other types of device.
Requirements:
- nullb_blk loadable kernel module
- configfs either as loadable module or built-in
The script follows common command line interface when there are multiple subcommands provided by one script. Currently there is:
- nullb setup - load required modules and mount configfs
- nullb create [-s sizeMB] [-z zonesizeMB] [-c zonecapacityMB] - create a new device with given sizes, allocating the first free indes, device is /dev/nullb$index
- nullb ls
- nullb list - show table of created null_blk devices, size and zone sizes
- nullb rm NAME
- nullb delete NAME - delete existing null_blk device by name, must match the device node name like nullb0
Defaults:
- device size: 2GiB
- zone size: 256MiB
# ./nullb setup
INFO: setup module and mounts
If that succeeds, the module has been loaded and the configfs is available.
# ./nullb create
INFO: Create nullb0
INFO: name=nullb0
INFO: size=2048M zone_size=256M
/dev/nullb0
The INFO messages are for clarity when run interactively, for scripting the last line contains the name of the device node that's been created.
# ./nullb ls
INFO: device nodes:
/dev/nullb0
INFO: created devices:
No Name Device Size Zone size Zone capacity On Mount
0 nullb0 /dev/nullb0 2048M 256M 200M 1 /mnt/test
The null block device nodes and the actual name in sysfs can be different so both are printed, though this script keeps both names respecting the index and name matches the device node filename.
# ./nullb rm nullb0
INFO: check mounts
INFO: removing nullb0
Mount checks just lists all mounts using the device but won't stop the removal. It should fail if the device is in use, otherwise the device and its resources are released after the command ends.
- the
null_blk
driver exists since kernel 3.12, zoned support has been added in 4.19 but please use at least 5.10 as a good base with improvements and fixes - the device data are backed by memory, reasonable size should be used for the 'size'
- zone sizes must be a power of two, eg. 4MiB works
- you can use the 'nullb' devices with 'btrfs' and have fun testing
- use
blkzone
utility to 'report' or 'reset' zones on the device - https://zonedstorage.io