Hacker News Reader

Comment by SomeCallMeTim | Hacker News Reader

In the case of Linux, Linus has a public position [1] that:

* A kernel module is run by the kernel, not incorporated in it,

* A kernel module only uses LGPL interfaces, so full GPL compliance shouldn't be necessary,

* Binary modules should be legal (though he doesn't like them), and

* Projects (like filesystems developed for other platforms) with their own life outside of Linux shouldn't be required to re-license if they clearly weren't based on Linux.

I know Linus doesn't represent all kernel developers, but his opinion should carry a lot of weight.

The SFConservancy is just practicing GPL overreach here, and is rattling a saber in hopes that Oracle blinks. Ubuntu has been distributing NVidia binary kernel modules for years and this hasn't resulted in a lawsuit, so it seems like an empty threat.

[1] http://linuxmafia.com/faq/Kernel/proprietary-kernel-modules....

View on HN

> The SFConservancy is just practicing GPL overreach here, and is rattling a saber in hopes that Oracle blinks.

Why would Oracle care? Oracle is not involved in OpenZFS and has closed source all of their changes for a while. The new code (which is a lot) is not owned by Oracle. It is owned by the individual contributors as OpenZFS is not doing copyright assignments.

> > The SFConservancy is just practicing GPL overreach here, and is rattling a saber in hopes that Oracle blinks.

> Why would Oracle care? Oracle is not involved in OpenZFS and has closed source all of their changes for a while. The new code (which is a lot) is not owned by Oracle. It is owned by the individual contributors as OpenZFS is not doing copyright assignments.

That's not how copyright works, and we've all seen how zealous Oracle loves to get with copyright arguments (the Android thing, for example).

That's exactly how copyright works. Without an assignment or contract, the author owns the copyright.

This probably doesn't affect the argument one way or the other, but in Germany, if some company employs you, and you write some code as part of your work for that company, AFAIK, copyright for that code automatically belongs to the company.

There are even cases where copyright for code you write after leaving the company could be considered to belong to that company.

The basic argument is that the OpenZFS folks have not been assigning their copyrights to Oracle after the Sun buyout and are not subject to what you have encountered in Germany (which I find just wrong, sorry for your troubles). Perhaps LGPL would be a better license for an OS (might be an interesting discussion).

> sorry for your troubles

I might have chosen my words poorly.

I have not had any troubles with this legal situation. When I am employed at a company that pays me to write code for them, I have no problem with the company owning the copyright.

And none of my former employers have ever tried to claim copyright on any code I have written once I stopped working there. Actually, I do not know of any specific case where this has happened.

All ZFS contributions before the illumos split were assigned to Oracle because they had a CLA with all contributors. I'd still argue they definitely have copyright ownership (the current source is based on the original files, no doubt about it).

How would Oracle own the code written without the CLA? That doesn't make any sense.

They don't own the current code, but all contributions are based on diffs on the original code. The contributors own their changes, but not the code as a whole. I don't believe that it's ever been tested in court how many incremental changes are enough to remove the original copyright owners' copyright claims. Especially when it comes to something like ZFS where there are data structures that were designed a long time ago and they must be kept (as well as the code for handling them) for backwards compatibility essentially indefinitely.

The issue isn't removing the original copyright holders claim. No one is saying that's possible. Oracle is free to relicense the code they own at any time, but that relicense doesn't affect the code written by others and Oracle does not have rights to that code unless they had a CLA. The new OpenZFS code is not under a CLA.

> his opinion should carry a lot of weight

That's nice, but it's only one side. What about the opinion of the developers who wrote the module and licensed it under CDDL?

The incompatibility between CDDL and GPL is well-known, and the authors at Sun made it that way deliberately.

It's not kosher to just say, "Well the authors of the GPL software are okay with it, so it's fine." You need to respect the wishes of the developers who contributed the CDDL code to begin with.

Something tells me it wasn't the developers at Sun who chose the CDDL for ZFS.

There's evidence there was no intent:

* https://news.ycombinator.com/item?id=11176361

* https://news.ycombinator.com/item?id=11176339

Well Danese Cooper, who wrote the thing, said she was deliberately made to author it to be incompatible.

I don't understand this line of reasoning, where somebody can say, "person A at Sun says it was intentional", then somebody else says, "person B at Sun says it wasn't intentional", and then acts as if that settles the matter.

All it means is that there's contention about it at best.

... and whether it was intentional or not doesn't affect what the text actually says.

The other reply [1] nails the real point: No one at Oracle is complaining. If they were, then one could care about their POV.

You also ignore the binary kernel object point: Nvidia's drivers are completely proprietary and closed source. How is that in any way less of a violation?

[1] https://news.ycombinator.com/item?id=11177128

There's lots to respond to in this comment, but I'm going to go away and come back to see if anyone else steps in to address them.

I don't want to seem like I'm trying to dominate the conversation.

(In the meantime, please see my other comments.)

> If they were, then one could care about their POV.

Oracle isn't exactly known for their attempts at fostering good-will with the open source community. Considering that, I don't think they have much incentive to complain now, even if their lawyers are 100% sure that there is a violation of their copyright.

Instead they can just wait for Canonical to hang themselves, then sue them and their customers for the fair market value plus damages. (Maybe they'll offer Canonical's customers a good deal when they switch to Oracle Linux.)

Sue them for what? No one anywhere has suggested even a theory where Oracle's license on ZFS is violated.

I thought TFA argues releasing a binary combining code under GPL 2 and CDDL 1 violates both of those licenses.

But my point was: IF Oracle thinks their copyright is being violated, they wouldn't complain NOW. If they don't think so, they wouldn't complain either. So Oracle staying silent shouldn't be taken as a signal of their opinion on the matter.

Canonical doesn't ship binaries containing ZFS+Linux. They ship the source code with scripts that build and insert it into the running kernel.

I believe that had been the case for some time, but it will no longer be true with the next Ubuntu release. (that is why everyone is particularly concerned about the issue right now)

If you look at the .diff.gz associated with Xenial's kernel, linked from http://packages.ubuntu.com/xenial/linux-image-4.4.0-7-generi... you will see that this diff adds spl/ and zfs/ directories to the linux source tree. I believe, but don't know for certain because the "[list of files]" links are all broken at the time I write this message, that zfs.ko is in the same binary .deb package as the GPL'd kernel.

That's not to say that mere aggregation is enough to trigger the portions of the GPL that also require a derived work in the copyright sense, but the Conservancy post we started with lays out in detail their view: A particular zfs.ko does nothing until it is linked with a matching version of the Linux kernel; the fact that the linking is dynamic and not static is irrelevant in their analysis.

>The other reply [1] nails the real point: No one at Oracle is complaining.

Canonical hasn't shipped their distro with the ZFS kernel module yet. My impression of Oracle is that they let their lawsuits do the talking.

>You also ignore the binary kernel object point: Nvidia's drivers are completely proprietary and closed source. How is that in any way less of a violation?

AFAIK NVidia's drivers are not a default part of any distribution (including Ubuntu from Canonical) and has to be enabled/installed explicitly by the end user, meanwhile Canonical said they were going to make the ZFS kernel module part of their default distribution.

Yes, I'm sure that Sun's legal department went through the trouble of creating a brand new license, and never checked whether it was incompatible with the GPL... despite the fact that their biggest competitor in the market was GPL licensed. Because that interpretation makes sense and isn't at all laughably naive.

The owners of CDDL-licensed code aren't complaining about a CDDL violation. It's only certain GPL advocates and licensors complaining of a GPL violation.

Certainly Oracle, Sun's legal successor, hasn't publicly complained.

> The owners of CDDL-licensed code aren't complaining about a CDDL violation

That may be true for the authors of ZFS (and I don't know that it is, but we can agree to accept that it might be true). There are people, though, that when looking for a license to use when releasing their software specifically choose one that is known to be incompatible with the GPL. Among them are people who go with the CDDL.

Saying that we should ignore the license text and treat the CDDL as compatible affects more than just the authors of the project we're talking about.

And if the authors of the code in question are really okay with it being combined with the GPL, then at any time they can make that explicit offering it under a license that is widely accepted to be compatible.

Are you arguing that the licensing requirements of the CDDL are the problem here? To the best of my knowledge (though I am not a lawyer and this is not legal advice) that is not the case.

My understanding is that what Canonical is doing here is not a violation of any CDDL licensing terms, and the legal question is purely of whether or not it is a violation of GPL license terms.

> That may be true for the authors of ZFS

Who are the only ones who have a legal interest in the manner in which the CDDL-licensed code at issue is used.

The fact that other people issue code with a license with the same text has no bearing on anything.

> Saying that we should ignore the license text and treat the CDDL as compatible affects more than just the authors of the project we're talking about.

No one is saying that, the argument is that this is not legally a derived work of the GPL-licensed work, so that neither the GPL nor compatibility between the CDDL and the GPL is relevant. No one involved has argued that the CDDL is, or should be treated as, compatible with the GPL.

The CDDL-licensed code is being distributed under the CDDL alone, not the GPL. That's the whole reason some people on the GPL side are complaining. I can't even see what people on the CDDL side would have to complain about under any interpretation of copyright law or either license that anyone has publicly made ever.

> There are people, though, that when looking for a license to use when releasing their software specifically choose one that is known to be incompatible with the GPL.

The issue at hand is not that the CDDL licensed software is being re-distributed under a GPL license. It's not. The issue is whether or not GPL software and CDDL software can interact in such a way as to violate the GPL license.

I doubt very much that "this can't be ported to a Linux kernel module" was a deciding factor when choosing the CDDL. If it was such an important factor, then they should have modified the CDDL to add such language to the license, which would make this issue much more clear.

I doubt very much that "this can't be ported to a Linux kernel module" was a deciding factor when choosing the CDDL.

The fly in that ointment is Danise Cooper's claim that they chose to base the CDDL on Mozilla's MPL 1.0 license in part for its GPL-incompatibility.

On the other hand, Sun's "Chief Open Source Officer", who introduced her at that talk as "the person who actually wrote the CDDL", later wrote she said that out of spite for losing the argument for licensing OpenSolaris as GPL.

So, who knows.

> There are people, though, that when looking for a license to use when releasing their software specifically choose one that is known to be incompatible with the GPL. Among them are people who go with the CDDL.

Are there really? Are there any who state so publicly? Or is this a purely theoretical concern?

Good point. I don't think I've ever heard of CDDL ever being used except by Sun (and Oracle as Sun's successor), and people redistributing work (or derivatives of work) originally released by Sun/Oracle.

I would imagine that the decision to use CDDL by anyone other than Sun/Oracle is based almost completely on "its the only license under which Sun/Oracle will allow me to release this work".

You will find that most fans of the CDDL are bigger fans of BSD licensing (e.g. node.js is BSD licensed), but there are a couple examples.

Offhand, I know that star and cdrtools are under a variant of the CDDL called CDDL-Schily, which is the CDDL with an addendum that software under it is "governed by the laws of Germany".

star and cdrtools are developed by someone who really loves Solaris and wants to make Linux more like it. Relicensing then under CDDL is just part of it; cdrtools also use Solaris-style device names even under Linux and he got quite angry with distros for patching it to use Linux's native device naming.

> The incompatibility between CDDL and GPL is well-known, and the authors at Sun made it that way deliberately.

To clarify, by "authors" do you mean individual developers? Or do you mean Sun/Oracle as a corporation.

If you mean individuals, how often do you see individual developers working for multinational companies just wing it with a license choice? (If that is what you are implying). Your comment made it seem as if CDDL was a long and hard personal choice made by each developers, and somehow people here are wiling to discount that hard decision.

I worked for 2 big companies and in neither case had a much of a personal choice in picking licenses for the code. As in thinking "hmm, I think this code today will be MIT, maybe tomorrow I'll feel like Public Domain". It is usually the higher ups / company owners / legal department that decides what the license is and the company as an entity owns the code.

Point being ownership of the code is very concentrated with ZFS, and in this case concentrated in the hands of Oracle + a few outside developers and it is just a file system (albeit a cool one). So that is way of the quickest solution to the impasse is for Oracle to re-license the code.

Of course the other fly in the ointment here is that Oracle has some stake in BTRFS -- a competing filesystem. It was project sponsored by them initially and so I think besides being, well, you know Oracle, they could be interested in stiffing adoption of ZFS to avoid competition with their own project.

> What about the opinion of the developers who wrote the module and licensed it under CDDL?

We are okay with it.

Although ZFS developers on all open source platforms are considered OpenZFS developers, I am explicitly listed on the wiki:

http://open-zfs.org/wiki/Contributors#Richard_Yao

I am also the #2 ZFSOnLinux contributor by commit count according to github's statistics and I have maintained that position for years:

https://github.com/zfsonlinux/zfs/graphs/contributors

If any of the other OpenZFS developers had a problem with Linux distributions shipping the OpenZFS code, I ought to have heard about that by now. So far, every OpenZFS developer on every other platform that I meet is happy that the code is available on Linux. I have even met employees at Oracle that are happy about it. While that is not an official stance of Oracle, Oracle itself publishes a dtrace module under the CDDL for use on Linux, so it would be strange to think that they would have a problem with shipping a port of the ZFS driver from the original OpenSolaris codebase under the original license for Linux, especially when it benefits Oracle Linux.

You're not one of the original copyright holders.. Oracle is one of the largest copyright holders. Maybe you should stop posting legal opinions when you clearly have your own bias?

I hold copyright on a significant enough amount of the Linux port that the license cannot easily change without my consent. The same is true for at least a dozen other people/organizations, if not a few dozen. You cannot say the same about yourself.

That being said, I am actively avoiding any statements that are repetitive with other comments that I have made on hacker news because they are not productive.

Exactly, you don't own enough of the code to change the license nor do you have the power to just magically change the license texts to allow them to become compatible. We can't allow this one violation without allowing all.

You assume that there is a violation, but I do not know of anyone who owns copyright on any portion of the ZFSOnLinux codebase that thinks that is the case. Canonical does not believe it either. Someone will need to convince us that there is a violation in order for any of us to try doing anything about it.

Also, your account was clearly registered in order to reply exclusively to me. Would you care to provide your actual name in a way that can be verified? Otherwise, you are just wasting my time, which will not last much longer as my only replies are going to be for you to identify yourself.

Someone emailed me to say that someone that Oracle might be able to change the terms of the CDDL for all software licensed under the CDDL as per sections 4.1 and 4.2. My discussions with lawyers thus far had focused on whether Oracle could change the license for just the OpenZFS code base (the answer is no) without any other copyright holders agreeing. Whether Oracle can change the definition of the CDDL for all software licensed under it (without the statement "version 1.0 only") is something that I need to run past a lawyer, but that might be possible.

I did not say it earlier, but this loophole would theoretically allow Oracle to publish a new version of the CDDL that is not open source at all and all software under the CDDL without a version peg would be avaliable under it.

I just consulted with another OpenZFS developer. All new files will be CDDL version 1.0 only. Consequently, this loophole is being closed.

Why can't new code be dual licensed, say, under the CDDL and the MIT license?

The CDDL has a copyleft provision, similar to the MPL. One of the clauses states "The Modifications that You create or to which You contribute are governed by the terms of this License."

The CDDL is file-based, just like the MPL that it was forked from, and was specifically authored to allow mixing with proprietary code. The only thing that the part you quoted means is that changes to existing files must be available under CDDL terms (or freer).

In any case, the contributors who own the rights to the code they submit are free to make it available under any other license they want, whether it be MIT or something else.

> > his opinion should carry a lot of weight

> That's nice, but it's only one side. What about the opinion of the developers who wrote the module and licensed it under CDDL?

The developers' opinions don't matter. All contributors used to have to sign a CLA to give copyright to Sun. Then Sun was bought by Oracle. So you should care about what Oracle thinks about copyright infringement.

> The incompatibility between CDDL and GPL is well-known, and the authors at Sun made it that way deliberately.

No, that's bullshit. CDDL (which is just the MPL) is a file-based copyleft because that's what license fitted their requirements. It's not their fault that people have a completely fucked up interpretation of "derivative work" that includes any combination of two completely separately developed works.

> It's not their fault that people have a completely fucked up interpretation of "derivative work" that includes any combination of two completely separately developed works.

Indeed, "derivative work" is a well-understood term of art in Copyright law. So the courts will have to decide if that's truly what they will interpret, or will they take the redistribution & usage restrictions of the GPL as a separable matter.

Oh please, even the author of the CDDL license states that this is so (Danese Cooper).

And it's not as if that admission was ever needed, were you even around back then ?

Solaris was being killed in the market by Linux, Sun decided to go open source in an effort to better compete with Linux, but there was NO way that they would give away their prized technology (ZFS, DTrace) under a license which would allow their main competitor to whom they were losing, to integrate said technology.

Hence CDDL, GPLv2 incompatible.

> Oh please, even the author of the CDDL license states that this is so (Danese Cooper).

Wikipedia mentions several other authors. Maybe Wikipedia is wrong, but I am doubtful only one person wrote the CDDL.

> Solaris was being killed in the market by Linux, Sun decided to go open source in an effort to better compete with Linux, but there was NO way that they would give away their prized technology (ZFS, DTrace) under a license which would allow their main competitor to whom they were losing, to integrate said technology.

That is one telling of the story. Other Sun employees have different tellings. It's not as cut and dry as you claim.

CDDL is GPL incompatible in a very subtle way. The copyright of the CDDL licensed code isn't infringed, it's technically the copyright of the GPL code that's infringed. CDDL allows distribution of the binaries under a different license (provided the license doesn't restrict the users' rights, which the GPL doesn't), but the source files must always be under the same license. The GPL requires the source files for the "dervied work" be under the same license, which is where the incompatibility steps in.

> The developers' opinions don't matter

I don't see how that's supposed to make sense. The developers/their employers retain copyright to their contributions—even those who signed the contributor agreement. (And it's dead, by the way. Nobody new is signing that agreement, and there's a backlog containing years of changes where the agreement doesn't apply.) So in a world where Oracle is making no indication of changing those terms, the developers' opinions do matter.

And I'm very specifically not just talking about the ZFS people here. This isn't "the ZFS license"; it's the CDDL. There's other software released under the CDDL.

> No, that's bullshit. CDDL (which is just the MPL) is a file-based copyleft because that's what license fitted their requirements. It's not their fault that people have a completely fucked up interpretation of "derivative work" that includes any combination of two completely separately developed works.

Again, this is completely incoherent to me. I recognize the words you're using, and I see that they're arranged in valid sentence structure, but I have no idea what point is supposed to lie in it or what it has to do with what I wrote.

And as I wrote to you before, the CDDL is not "just the MPL". Sun didn't just rename it and appoint themselves as license stewards when they forked the MPL. They made significant changes. They are not the same license.

Reply on news.ycombinator.com