nvmetcfg is a tool to configure the NVMe Target (nvmet) subsystem for Linux.

• Simple CLI commands with friendly help texts and error messages.
• Show and configure Ports, Subsystems and Namespaces.
• Strict checks to ensure only valid configuration gets applied.
• Save the state in a simple yaml file and restore it later.
• Full integration tests and unit tests ensure it functions as advertised.

nvmetcfg primarily provides the nvmet CLI tool. If you provide no arguments, you will see the main usage and subcommands:

NVMe-oF Target Configuration CLI

Usage: nvmet <COMMAND>

Commands:
  port       NVMe-oF Target Port Commands
  subsystem  NVMe-oF Target Subsystem Commands
  namespace  NVMe-oF Target Subsystem Namespace Commands
  state      NVMe-oF Target Subsystem State Management Commands
  help       Print this message or the help of the given subcommand(s)

Options:
  -h, --help  Print help

This should be familiar to anyone who has used clap-based CLIs before.

Each of the subcommands contains a usage as well. When in doubt, run help <subcommands>. Don't know what arguments to provide to port add? Run help port add.

Keep in mind that you need at least the nvmet module loaded. Given that this tool is modifying the kernel sysfs, manipulating the state requires running as root.

Alternatively, this project also provides a library for integration into other projects. In this case, consider the nvmet binary source code in src/bin/nvmet/ as the example.

This is a simple example showing:

• Creation of a 1G file /tmp/test.img.
• Setup of /dev/loop0 to point to /tmp/test.img
• Creation of a simple single-namespace subsystem nqn.2023-11.sh.tty:example-test-loop which is backed by /dev/loop0.
• Creation of a NVMe over TCP port which binds to all IPv4 addresses on port 4420, the standard port.
• Saving the current nvmet subsystem state, clearing it and restoring it.

# whoami
root
# modprobe nvmet 
# fallocate -l 1G /tmp/test.img
# losetup /dev/loop0 /tmp/test.img
# nvmet subsystem add nqn.2023-11.sh.tty:example-test-loop
# nvmet namespace add nqn.2023-11.sh.tty:example-test-loop 1 /dev/loop0
# nvmet subsystem show
Configured subsystems: 1
Subsystem: nqn.2023-11.sh.tty:example-test-loop
	Allow Any Host: true
	Number of Namespaces: 1
	Namespaces: 1
# nvmet namespace show nqn.2023-11.sh.tty:example-test-loop
Number of Namespaces: 1
Namespace 1:
	Enabled: true
	Device Path: /dev/loop0
	Device UUID: 75db752f-9c96-4e3b-ae08-e0feebe08138
	Device NGUID: 00000000-0000-0000-0000-000000000000
# nvmet port add 1 tcp 0.0.0.0:4420
# nvmet port add-subsystem 1 nqn.2023-11.sh.tty:example-test-loop
# nvmet port show
Configured ports: 1
Port 1:
	Type: Tcp(0.0.0.0:4420)
	Subsystems: 1
		nqn.2023-11.sh.tty:example-test-loop

# mkdir -p /etc/nvmetcfg
# nvmet state save /etc/nvmetcfg/state.yaml
Sucessfully written current state to file.
# nvmet state clear
Sucessfully cleared configuration: 2 state changes.
# nvmet state restore /etc/nvmetcfg/state.yaml
Sucessfully applied saved state: 2 state changes.
# nvmet state restore /etc/nvmetcfg/state.yaml
No changes made: System state has no changes compared to saved state.

Obviously, the show commands are not necessary for functionality, only for visual verification. If any of the commands fail, error messages will be printed.

For an example of the config file, check out examples/tcp.yaml. It should match what you'd get if running this, other than the random serial number.

On NixOS, you can install the tool by adding github:vifino/nvmetcfg as a flake. It provides the well-known system-specific packages attribute, specifically .packages.${system}.nvmetcfg. Currently, the flake supports x86, x86_64 and aarch64 systems.

Alternatively, with NixOS or plain Nix on Linux, you can just temporarily run nvmet for testing:

modprobe nvmet
nix run github:vifino/nvmetcfg help

Of course, automatically loading the kernel modules is the way to go for actual usage:

{
  boot.kernelModules = [ "nvmet" ];
}

Just the usual lifecycle of Cargo-based Rust projects:

$ cargo build --release

Running it is simple:

$ sudo modprobe nvmet
$ sudo ./target/release/nvmet 

sudo is used here, but that's just to get the point across that you probably need to run this as root if you actually wanna modify the state.

nvmet-cli is a Python project that has been there since the beginning. It is written by a maintainer of the kernel nvmet subsystem itself and does the job.

However, there are some things that I (vifino) disliked:

• More or less a 1:1 mapping of config to sysfs instead of attempting to simplify redundancy.
• Instead of being a tool you mostly use within your shell, it has it's own interactive shell.
  • This makes not just scripting harder, but also quick state inspection.. clunky.
• The code and project structure seem suboptimal, in addition to being sparsely maintained.
• I don't like interpreted languages running as root and touching my kernel.

This project is my attempt to create a tool that is more like the incredible nvme-cli. It reached my goal of feature parity and integrates better in my workflows.