LXD now re-licensed and under a CLA

The facts

As of earlier today, right before the release of LXD 5.20, Canonical made a couple of changes to LXD which are sure to have a serious impact to LXD users and downstream projects that integrate with LXD or provide solutions based on it.

The first is the re-licensing of LXD from the Apache 2.0 license to the AGPLv3 license.
This happened in: https://github.com/canonical/lxd/pull/12663

The second is the addition of the Canonical CLA as requirement for all further contributions.
This happened in: https://github.com/canonical/lxd/pull/12665

Disclaimer

What’s below is my personal analysis of the situation, it is not legal advice, anyone affected is very strongly encouraged to seek proper legal advice from their counsel.

Real license of LXD

Per the commit message performing the re-licensing, all further contributions will be under the AGPLv3 license and all contributions from Canonical employees have been re-licensed to AGPLv3.

However, Canonical does not own the copyright on any contribution from non-employees, such as the many changes they have imported from Incus over the past few months. Those therefore remain under the Apache 2.0 license that they were contributed under.

As a result, LXD is now under a weird mix of Apache 2.0 and AGPLv3 with no clear metadata indicating what file or what part of each file is under one license or the other.

This is likely to make it very “fun” for anyone performing licensing reviews to evaluate LXD for adoption in their environment.

Impact to LXD users

For LXD users, other than potentially triggering corporate policies that ban the use of AGPLv3 software (more common than one may think), the impact should be minimal. It’s still the same LXD and it’s still open source software.

However, if you were altering LXD in any way, then you will need to familiarize yourself with the AGPLv3 license as unlike Apache 2.0, it does require any changes be made available under the AGPLv3 even if you don’t expose your users to your modified binaries. This is the main design characteristics of the AGPLv3 license, it was meant to force those operating modified versions of open source as a hosted service to share their modifications.

Impact to downstreams (consumer of LXD Go packages)

Up until now, all the Go packages of LXD were under the Apache 2.0 license, that was fitting quite well in the Go ecosystem where the Apache 2.0, BSDs and MIT licenses are very popular.

Now with this change, you need to realize that you may start to include/bundle AGPLv3 code within your own project. This a copyleft license and so may require re-licensing of your own project to comply with it.

Again, this is quite the can of worms, with my usual recommendation being “stay away”, but if you must use any of LXD’s Go packages, I’d strongly recommend talking to a lawyer to fully understand your exposure to that new license.

Impact on Incus

Now for what obviously impacts me the most, what this is going to do to Incus.

As a brief reminder Incus is a fork of LXD which was started in August 2023.
So far, it’s been tracking LXD changes, applying those that make sense and otherwise fixing bugs and making improvements of its own, as most forks do.

This change from Canonical is going to be causing two unfortunate side effects:

  • Incus will no longer be including changes originating from LXD as that would require us to include AGPLv3 code into our codebase and so get us into the same mixed license mess as LXD now put itself. This is obviously unacceptable to us, we very much like licensing clarity and quite enjoy the Apache 2.0 license.
  • LXD will similarly no longer be able to take changes from Incus, as those are going to remain under the Apache 2.0 license and more importantly, will not have been released under the Canonical CLA.

To enforce that second part, the tooling we’ve been using thus far to monitor LXD changes and automatically backport them to Incus will be used to detect any changes to LXD which originated from Incus. Unless the author gave express consent for them to be released under a different license and under the Canonical CLA, those changes should not be included in LXD.

Incus is also a consumer of the LXD Go API in the lxd-to-incus tool. Thankfully, we have no need for anything recent in there, so will simply be making sure that we never import code past the licensing change.

Conclusion

Overall, I’m very disappointed, although absolutely not surprised in seeing this change happen.
It’s certainly going to be quite annoying for Incus, and I suspect this is the whole point of it.

But it’s also a very odd move by Canonical as it puts LXD into a problematic grey area as far as its true license is concerned which will likely seriously hurt its adoption both by companies and distributions.

In any case, I’d urge anyone who has concerns about this change to reach out to their legal representation and maybe consider switching over to Incus where we will happily keep releasing our CLA-free Apache 2.0 licensed fork of this once great project.

Update (2023/12/12)

I’ve seen a number of people point out that Apache 2.0 is compatible with the AGPLv3 and that’s certainly correct, however compatibility doesn’t mean that the code in question suddenly becomes AGPLv3, it means that it can be included in an AGPLv3 project. So I’d still expect there to be good tracking of the license of the individual files / code chunks so that someone can tell whether a particular piece of code is AGPLv3 or Apache 2.0, this is currently not possible.

The code mentions headers (probably SPDX) to be present whenever code isn’t AGPLv3, but no such headers were introduced at time of writing.

The announcement also very specifically spells out that past contributions are not being re-licensed and therefore remain under the Apache 2.0 license, though again, there is currently no way to identify what contributions that is. So this still leads to LXD now being a mix of AGPLv3 and Apache 2.0 with no way to figure out which is which.

All that’s known for sure is that all new contributions are to be under AGPLv3 and must be from copyright holders (author or employer) who has signed the Canonical CLA. Those two will preclude the inclusion of any Incus code in LXD moving forward.

Update (2023/12/14)

Canonical now seems to be trying to get around the fact that Incus isn’t AGPLv3 by marking the code they import from us as Apache 2.0 in the commit message. While this is definitely fine from a licensing standpoint given the compatibility between AGPLv3 and Apache 2.0 license, it still violates their requirement that all new contributions be under the Canonical CLA.

About Stéphane Graber

Project leader of Linux Containers, Linux hacker, Ubuntu core developer, conference organizer and speaker.
This entry was posted in Incus, Planet Ubuntu, Zabbly. Bookmark the permalink.

11 Responses to LXD now re-licensed and under a CLA

  1. Linux User 123 says:

    Seriously, don’t sign a CLA:
    https://drewdevault.com/2023/07/04/Dont-sign-a-CLA-2.html

  2. Interesting. I suppose this means in a nutshell that Incus is a hard fork.

    1. Yep, that is forcing Incus into being a hard fork now.

      1. Simon Klyne says:

        Fork Yeah! (quoting Brian Cantrill when Oracle forced Illumos to hard fork)

  3. klh says:

    Never thought I would be against someone using one of the GPL licenses. Absolutely pathetic move by Canonical.

  4. Linus T. Fan says:

    I’ve been manually building lxd source code for years on Raspbian and Ubuntu 22.04 to avoid the gross snap stuff… and now when the native lxd binary works well on Debian Bookworm with simple “apt install”, which is fantastic, then Canonical starts doing this…
    We already switch from certbot to acme.sh, and we never mind to do this again from lxd to incus.

  5. Nino says:

    Although the AGPL is my favorite license (I like the fact that it prohibits large companies from unfairly profiting from libre projects) the impact on Incus (and maybe LXD itself) is sad.
    I understand that using AGPL (or GPL for that matter) add a barrier to the adoption of projects such as Incus and LXD by companies but I think it’s generally better than providing free work for tech giants and hope they will upstream fixes & improvements.

    However, for such change in an existing project, you have to take the ecosystem into account. Here, in the ecosystem, we had Incus for example, a friendly fork that is forced to change to a hard fork.. It’s seems like a waste of energy and contributors for both project.

    Indeed that change coupled with the new CLA requirements is not very outgoing towards community..

  6. foo says:

    You may unironically consider switching to The JSON License or some other license that is not compatible with AGPL.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.