nvme-cli

NVM-Express user space tooling for Linux.

Build from source

nvme-cli uses meson as build system.

nvme-cli dependencies:

Configuring

In case libnvme is not installed on the system, it possible to use meson's fallback feature to resolve the dependency.

$ meson setup --force-fallback-for=libnvme .build

If the libnvme is already installed on the system meson is using pkg-config to find the dependency. In this case a plain setup call is enough:

$ meson setup .build

With meson's --wrap-mode argument it's possible to control if the additional dependencies should also resolved or not. The options are

--wrap-mode {default,nofallback,nodownload,forcefallback,nopromote}

Note for nvme-cli the 'default' is set to nofallback.

Building

$ meson compile -C .build

Installing

# meson install -C .build

Makefile wrapper

There is a Makefile wrapper for meson for backwards compatibility

$ make
# make install

Note in this case libnvme needs to be installed by hand first.

RPM build support via Makefile that uses meson

$ make rpm

If not sure how to use, find the top-level documentation with:

$ man nvme

Or find a short summary with:

$ nvme help

Distro Support

Many popular distributions (Alpine, Arch, Debian, Fedora, FreeBSD, Gentoo, Ubuntu, Nix(OS), openSUSE, ...) and the usual package name is nvme-cli.

OpenEmbedded/Yocto

An nvme-cli recipe is available as part of the meta-openembeded layer collection.

Buildroot

nvme-cli is available as buildroot package. The package is named nvme.

Developers

You may wish to add a new command or possibly an entirely new plug-in for some special extension outside the spec.

This project provides macros that help generate the code for you. If you're interested in how that works, it is very similar to how trace events are created by Linux kernel's 'ftrace' component.

Add command to existing built-in

The first thing to do is define a new command entry in the command list. This is declared in nvme-builtin.h. Simply append a new "ENTRY" into the list. The ENTRY normally takes three arguments: the "name" of the subcommand (this is what the user will type at the command line to invoke your command), a short help description of what your command does, and the name of the function callback that you're going to write. Additionally, You can declare an alias name of subcommand with fourth argument, if needed.

After the ENTRY is defined, you need to implement the callback. It takes four arguments: argc, argv, the command structure associated with the callback, and the plug-in structure that contains that command. The prototype looks like this:

int f(int argc, char **argv, struct command *cmd, struct plugin *plugin);

The argc and argv are adjusted from the command line arguments to start after the sub-command. So if the command line is "nvme foo --option=bar", the argc is 1 and argv starts at "--option".

You can then define argument parsing for your sub-command's specific options then do some command specific action in your callback.

Add a new plugin

The nvme-cli provides macros to make define a new plug-in simpler. You can certainly do all this by hand if you want, but it should be easier to get going using the macros. To start, first create a header file to define your plugin. This is where you will give your plugin a name, description, and define all the sub-commands your plugin implements.

There is a very important order on how to define the plugin. The following is a basic example on how to start this:

File: foo-plugin.h

#undef CMD_INC_FILE
#define CMD_INC_FILE plugins/foo/foo-plugin

#if !defined(FOO) || defined(CMD_HEADER_MULTI_READ)
#define FOO

#include "cmd.h"

PLUGIN(NAME("foo", "Foo plugin"),
	COMMAND_LIST(
		ENTRY("bar", "foo bar", bar)
		ENTRY("baz", "foo baz", baz)
		ENTRY("qux", "foo quz", qux)
	)
);

#endif

#include "define_cmd.h"

In order to have the compiler generate the plugin through the xmacro expansion, you need to include this header in your source file, with pre-defining macro directive to create the commands.

To get started from the above example, we just need to define "CREATE_CMD" and include the header:

File: foo-plugin.c

#include "nvme.h"

#define CREATE_CMD
#include "foo-plugin.h"

After that, you just need to implement the functions you defined in each ENTRY, then append the object file name to the meson.build "sources".

meson tips

In case meson doesn't find libnvme header files (via pkg-config) it will fallback using subprojects. meson checks out libnvme in subprojects directory as git tree once to the commit level specified in the libnvme.wrap file revision parm. After this initial checkout, the libnvme code level will not change unless explicitly told. That means if the current branch is updated via git, the subprojects/libnvme branch will not updated accordingly. To update it, either use the normal git operations or the command:

$ meson subprojects update

Dependency

libnvme depends on the /sys/class/nvme-subsystem interface which was introduced in the Linux kernel release v4.15. Hence nvme-cli 2.x is only working on kernels >= v4.15. For older kernels nvme-cli 1.x is recommended to be used.

How to contribute

There are two ways to send code changes to the project. The first one is by sending the changes to [email protected]. The second one is by posting a pull request on github. In both cases please follow the Linux contributions guidelines as documented in

https://docs.kernel.org/process/submitting-patches.html#

That means the changes should be a clean series (no merges should be present in a github PR for example) and every commit should build.

See also https://opensource.com/article/19/7/create-pull-request-github

How to cleanup your series before creating PR

This example here assumes, the changes are in a branch called fix-something, which branched away from master in the past. In the meantime the upstream project has changed, hence the fix-something branch is not based on the current HEAD. Before posting the PR, the branch should be rebased on the current HEAD and retest everything.

For example rebasing can be done by following steps

# Update master branch
#   upstream == https://github.com/linux-nvme/nvme-cli.git
$ git switch master
$ git fetch --all
$ git reset --hard upstream/master

# Make sure all dependencies are up to date and make a sanity build
$ meson subprojects update
$ ninja -C .build

# Go back to the fix-something branch
$ git switch fix-something

# Rebase it to the current HEAD
$ git rebase master
[fixup all merge conflicts]
[retest]

# Push your changes to github and trigger a PR
$ git push -u origin fix-something

Persistent, volatile configuration

Persistent configurations can be stored in two different locations: either in the file /etc/nvme/discovery.conf using the old style, or in the file /etc/nvme/config.json using the new style.

On the other hand, volatile configurations, such as those obtained from third-party tools like nvme-stats or blktests' can be stored in the /run/nvme directory. When using the nvme-cli tool, all these configurations are combined into a single configuration that is used as input.

The volatile configuration is particularly useful for coordinating access to the global resources among various components. For example, when executing blktests for the FC transport, the nvme-cli udev rules can be triggered. To prevent interference with a test, blktests can create a JSON configuration file in /run/nvme to inform nvme-cli that it should not perform any actions trigged from the udev context. This behavior can be controlled using the --context argument.

For example a blktests volatile configuration could look like:

[
  {
    "hostnqn": "nqn.2014-08.org.nvmexpress:uuid:242d4a24-2484-4a80-8234-d0169409c5e8",
    "hostid": "242d4a24-2484-4a80-8234-d0169409c5e8",
    "subsystems": [
      {
	"application": "blktests",
        "nqn": "blktests-subsystem-1",
        "ports": [
          {
            "transport": "fc",
	    "traddr": "nn-0x10001100aa000001:pn-0x20001100aa000001",
	    "host_traddr": "nn-0x10001100aa000002:pn-0x20001100aa000002"
          }
        ]
      }
    ]
  }
]

Note when updating the volatile configuration during runtime, it should done in a an atomic way. For example create a temporary file without the .json file extension in /run/nvme and write the contents to this file. When finished use rename to add the '.json' file name extension. This ensures nvme-cli only sees the complete file.