SmartOS (based on IllumOS) released

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 16:26 UTC (Mon) by gidoca (subscriber, #62438)
In reply to: SmartOS (based on IllumOS) released - with KVM by kragilkragil2
Parent article: SmartOS (based on IllumOS) released - with KVM

Looks like they don't consider kernel modules derived work, see here.

to post comments

Modules

Posted Aug 15, 2011 16:29 UTC (Mon) by corbet (editor, #1) [Link] (3 responses)

OK, that's useful. One has to agree that loading KVM as a module into a Solaris kernel cannot turn the kernel into a derived work of KVM - as long as they didn't put KVM-derived stuff there already to make that loading possible. It seems that they are saying that this module is also not a derived work of the Solaris kernel, so the CDDL's requirements won't apply to it.

Modules

Posted Aug 15, 2011 19:12 UTC (Mon) by trasz (guest, #45786) [Link] (2 responses)

It's not them who is claiming this - it's the CDDL license:

1.3. Covered Software means (a) the Original Software, or (b) Modifications, or (c) the combination of files containing Original Software with files containing Modifications, in each case including portions thereof.

1.9. Modifications means the Source Code and Executable form of any of the following:

A. Any file that results from an addition to, deletion from or modification of the contents of a file containing Original Software or previous Modifications;

B. Any new file that contains any part of the Original Software or previous Modification; or

C. Any new file that is contributed or otherwise made available under the terms of this License.

Since the KVM code does not fit to these claims, it's not covered by the CDDL license.

No breach of CDDL terms here

Posted Aug 22, 2011 23:26 UTC (Mon) by webmink (guest, #47180) [Link] (1 responses)

Since no CDDL-licensed file appears to be being modified or merged with KVM, I can see no reason at all why KVM needs to be covered by CDDL. Remember CDDL is a file-l;evel, weak copyleft license like the MPL (or indeed in some circumstances like LGPL) and thus merely loading KVM into the Illumos kernel does not create a need for KVM to be CDDL-licensed.

No breach of CDDL terms here

Posted Aug 23, 2011 1:17 UTC (Tue) by mjg59 (subscriber, #23239) [Link]

It's true that there's probably no breach of the CDDL here, but the GPL requires that the entire work be available under terms compatible with the GPL. If the final work is a derivative of both CDDLed work and GPLed work, then the CDDLed work also needs to be available under terms compatible with the GPL. The question of whether it's a derivative work of the CDDLed code is one that's determined by law and the courts, not by the CDDL.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 17:13 UTC (Mon) by fuhchee (guest, #40059) [Link] (25 responses)

"they don't consider kernel modules derived work"

Whether something is or is not a "derived work" for copyright purposes is not a matter of the author's, nor the license writer's preference. It comes from the law.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 18:17 UTC (Mon) by robert_s (subscriber, #42402) [Link] (11 responses)

"Whether something is or is not a "derived work" for copyright purposes is not a matter of the author's, nor the license writer's preference. It comes from the law."

Welcome to the Solaris community.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 18:40 UTC (Mon) by danieldk (guest, #27876) [Link] (10 responses)

License-wise (IllumOS/Joyent/Nexenta) they are stuck in a hard place, thanks to Sun. CDDL provides none of the advantages of non-copyleft licenses (easy re-use in proprietary software), nor benefits that the GPL would give (code sharing with Linux).

Of course, this was exactly Sun's intention, and it goes to show that picking non-standard licenses is not really helping the wider community.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:02 UTC (Mon) by trasz (guest, #45786) [Link] (9 responses)

Completely wrong. CDDL, differently from GPL, does not have the virality problem, so there is no problem with reusing the code in proprietary software. At the same time, it has all the advantages of copyleft license (it's just a slightly tweaked Mozilla license). Yes, it is impossible to share code with Linux, but that can be said about any license that is not GPL or subset of GPL.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:28 UTC (Mon) by danieldk (guest, #27876) [Link] (4 responses)

Completely wrong.

You are stating this as if there is no difference between the CDDL and non-copyleft licenses, which is not true. Any proprietary software vendor would have to release the source code to any CDDL-licensed file, including proprietary modifications.

Yes, it is impossible to share code with Linux, but that can be said about any license that is not GPL or subset of GPL.

Well, some of the other major licenses are compatible with the GPL, such as the modified BSD License and the X11 (MIT) license. Of course, the viral nature will taint, say a BSD operating system.

Anyway, the GPL has been around for ages and CDDL is relatively new. Pretty much everyone agrees that Sun did not use the GPL or a compatible license for the fear of losing an advantage in some areas. That was their right, but their contribution to the FLOSS community should not be overstated. They licensed it in such a way that the code is not really useful outside the OpenSolaris community.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:48 UTC (Mon) by trasz (guest, #45786) [Link] (3 responses)

Yes, proprietary software vendor has to release it's modifications to the CDDL code; that's the point of copyleft, after all. However, differently from GPL, at least in Stallman's interpretation, the vendor does not need to release _his_ code that is merely linked with CDDL-ed code.

As for the GPL compatibility - MIT and modified BSD are subset of GPL.

As for the usefulness: licensing OpenSolaris under GPL would make it useless to everyone except Linux community, and Linux developers probably wouldn't accept the code anyway due to NIH. By licensing it under CDDL they made the code available for pretty much everyone, including e.g. MacOS X, and even Linux: using CDDL-ed kernel module is not in any way more problematic than using closed source kernel module.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:58 UTC (Mon) by clump (subscriber, #27801) [Link] (2 responses)

As for the usefulness: licensing OpenSolaris under GPL would make it useless to everyone except Linux community, and Linux developers probably wouldn't accept the code anyway due to NIH. By licensing it under CDDL they made the code available for pretty much everyone, including e.g. MacOS X, and even Linux: using CDDL-ed kernel module is not in any way more problematic than using closed source kernel module.

You lost me here. The parent comment regarded the decision to conflict with the GPL, not about changing from CDDL to GPL. Your "NIH" comment is an unfortunate non-sequitur.

Binary modules are very problematic for Linux distributors and users. If the module is not binary but under a conflicting license the situation is slightly better, only in that you can see the code if you were so inclined to look. That still would not make maintenance, distribution, or a user's life any easier.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 20:13 UTC (Mon) by trasz (guest, #45786) [Link] (1 responses)

Ah, I assumed you were talking about releasing the code under GPL, not about making CDDL GPL-compatible. But again - the only way to make CDDL GPL-compatible would be to make it a subset of GPL, it couldn't include the nice patent protection it has now, for example.

As for binary modules - yes, they are a problem, but it's just an annoyance. Releasing the code under GPL instead of CDDL would make it impossible to e.g. include DTrace in OSX - not "inconvenient", but just impossible, at least according to Stallman's interpretation of GPL.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 22:25 UTC (Mon) by cortana (subscriber, #24596) [Link]

Surely the code could be dual-licensed under the GPL and the CDDL.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 9:04 UTC (Tue) by rvfh (guest, #31018) [Link] (3 responses)

(...) the virality problem (...)

Could you please refrain from using this word? You get viruses against your will and they try to destroy you, whereras you choose to derive from GPL code and I fail to see that destroy anybody.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 20:13 UTC (Tue) by jone (guest, #62596) [Link]

Actually, I think they misspelled virility .. while the code and ideas are quite high caliber, I've always felt that due to the CDDL [open]solaris have historically suffered from an impotence problem with a slight overcompensation by bragging about their 64 bit structures and large disk drives. :)

SmartOS (based on IllumOS) released - with KVM

Posted Aug 17, 2011 2:43 UTC (Wed) by rahvin (guest, #16953) [Link] (1 responses)

Just an unrelated aside. The implied evil of viruses is your own cultural predication. A properly evolved virus doesn't kill the host or really even harm them, in fact a virus that results in death is a virus that isn't very good at it's job. The human genome is littered with the DNA of Viruses from the past that got their DNA incorporated into our own permanently. If the virus kills the host they've eliminated a transmission vector and hurt their own survival prospect. A virus that's good at it's job succeeds in infecting every possible host without damage to the host, thus ensuring permanent survival of the DNA/RNA.

Viruses and disease are a natural evil in Human language but modern science has revealed the virus to be a very flexible organism. Some of those are beneficial and helpful. In the same vein a "viral" GPL isn't necessarily a bad thing, it just depends on perspective.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 17, 2011 21:35 UTC (Wed) by k8to (guest, #15413) [Link]

It's bad in that it's inaccurate.

Viruses propogate without permission from their host. That's very much what the phrase is used to imply is going on.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:06 UTC (Mon) by trasz (guest, #45786) [Link] (11 responses)

Actually, license can play a role here. CDDL, for example, clearly defines that things merely linked with the CDDL-ed code are not considered derived work, thus avoiding potential virality.

As for the Linux kernel modules - that's not clear either. It would be silly to consider e.g. nVidia drivers or the code VMWare ESX puts into the kernel a derived work.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:44 UTC (Mon) by fuhchee (guest, #40059) [Link] (4 responses)

"clearly defines that things merely linked with the CDDL-ed code are not considered derived work"

It doesn't matter what they clearly define. They don't own the definition, the law does. For works that are not "derived works" in the copyright law sense, the license is apprx. irrelevant.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:51 UTC (Mon) by trasz (guest, #45786) [Link] (2 responses)

It does matter: for works that are not derived work in the copyright law sense, the license is irrelevant, but for work that could be derived work, the license _is_ relevant. The problem with GPL is that the law does not state clearly whether stuff merely linked with GPL-ed code constitutes derived work; CDDL solves the problem by explicitly saying it does not apply to the "merely linked" stuff.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 17, 2011 15:09 UTC (Wed) by vonbrand (guest, #4458) [Link] (1 responses)

Exactly as murky as the GPL case. What the heck is "merely linked"? To find out what functions/data I can use from "the other side" presumably I have to study it. Whatever I write and compile to "merely link" won't be able to stand on it's own, and would be completely useless in the next operating system (or even a few versions later of the same). So the argument that it isn't derived is pretty thin. Sure, it could be non-derived (just as in the GPL case), but in most cases it will be a bona fide derived work.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 17, 2011 18:17 UTC (Wed) by trasz (guest, #45786) [Link]

The "merely linked" wording is my own. CDDL is very specific here, see my other comment above. It makes it very clear that it only applies to files copyrighted under CDDL or containing code taken from files copyrighted under CDDL. This makes it very different from GPL, which, according to RMS, applies also to everything linked with files copyrighted under GPL, i.e. the phenomenon often called "virality".

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 20:05 UTC (Mon) by allan1015 (guest, #78373) [Link]

>>>It doesn't matter what they clearly define. They don't own the definition, the law does

Ah yes as the last poster commented - that view is while ideal a bit naive as the 'law' lends absolutely no guidance on GPL derived works/copyleft.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 19:48 UTC (Mon) by clump (subscriber, #27801) [Link]

Similar to NVidia binary drivers in Linux, one concern would regard the permissibility of distributing binaries of conflicting licenses.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 15, 2011 22:18 UTC (Mon) by robert_s (subscriber, #42402) [Link] (2 responses)

"thus avoiding potential virality."

Do we have to have this "virality" guff? It's being used very spuriously. I've never known of a virus that requires you to voluntarily choose to use it before it infects you.

It just makes you look like you've got an agenda.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 20:21 UTC (Tue) by jone (guest, #62596) [Link]

perhaps that should read .. "thus avoiding potential virility" .. maybe trasz is just a bit jealous of the reproductive effectiveness the linux codebase has already had .. :)

SmartOS (based on IllumOS) released - with KVM

Posted Aug 18, 2011 14:42 UTC (Thu) by seyman (subscriber, #1172) [Link]

Agreed. When describing Linux's license, I use "transitivity" when talking to people with a scientific background and "hereditary" in all other cases. Both terms describe the GPL's effects far better than "virality"

SmartOS (based on IllumOS) released - with KVM

Posted Aug 27, 2011 3:35 UTC (Sat) by cas (guest, #52554) [Link] (1 responses)

As for the Linux kernel modules - that's not clear either. It would be silly to consider e.g. nVidia drivers or the code VMWare ESX puts into the kernel a derived work.

No, it wouldn't be silly to do that. In fact, that's precisely why proprietary drivers like nvidia's can not be distributed WITH the kernel. Combining the two produces a derived work.

That's something that an end-user is free to do on their own machines (since they're not distributing the combined result, the GPL's rules on redistribution of derived works does not apply. The GPL restricts redistribution, not usage.)

BUT - and here's the catch - a distributor (such as debian or redhat or anyone else) can not do that. They can not re-distribute the combined work under terms compatible with the GPL so they are not permitted to re-distribute it at all.

At best, they can create an installer package which makes it more convenient for an end-user to download, patch the kernel and/or compile the module, and install it. as, for example, debian does with the nvidia-kernel-dkms package in the debian non-free archive. This actually works quite well in practice and is no more inconvenient to the end-user than similar dkms packages for GPL-compatible kernel modules that aren't part of the mainline kernel - e.g. oss4-dkms, blcr-dkms, xtables-addons-dkms and others.

(btw, this is NOT a bug in the GPL. it's a feature. its purpose is to prevent hijacking GPL-ed code into proprietary software)

SmartOS (based on IllumOS) released - with KVM

Posted Aug 27, 2011 11:10 UTC (Sat) by fuhchee (guest, #40059) [Link]

"Combining the two produces a derived work."

But then the question becomes what's "combining" in this context: shipping on the same installation DVD? or installation server?

It's nice that most OSS companies tend to respect an expansive reading of all these issues, to avoid litigation risks and alienating developers. But don't mistake that for legal clarity of the situation.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 21:27 UTC (Tue) by vonbrand (guest, #4458) [Link]

The author (really the copyright owner) has no say in this, the law (really a court as the law itself is silent on this point) determines what a derived work is in such a murky case. If they load the Linux module into a modified Solaris kernel, maybe the modifications to Solaris are, perhaps they aren't.

The copyright owners could approve of this use, giving a special exception for this, in case the law says it is a derived work.

Or SmartOS could just play dumb and forge ahead expecting nobody with a standing will complain...

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 7:08 UTC (Tue) by rodgerd (guest, #58896) [Link] (11 responses)

"I've decided I can use GPL code because I've written my license to say it's OK and doesn't taint my codebase" would be called out as bullshit in no time flat if it was Microsoft; I see no reason why some random OpenSolaris derivative should get a free pass.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 9:54 UTC (Tue) by nye (guest, #51576) [Link] (10 responses)

>"I've decided I can use GPL code because I've written my license to say it's OK and doesn't taint my codebase" would be called out as bullshit in no time flat if it was Microsoft; I see no reason why some random OpenSolaris derivative should get a free pass

Because it's far from clear that a kernel should be legally considered a derivative work of some module that you build for it. In fact, their position sounds pretty safe.

Thought experiment: if somebody writes a kernel driver for Windows and releases it under the GPL does that mean we should 'call [it] out as bullshit'? Same thing.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 12:11 UTC (Tue) by pboddie (guest, #50784) [Link]

Because it's far from clear that a kernel should be legally considered a derivative work of some module that you build for it.

You have to work back from the objectives of the GPL to understand what the licence is trying to do. The aim of the GPL is to have a software system whose complete code exhibits the property of offering the four freedoms. When the licence talks about a derived work, it attempts to use established terminology to entertain the possibility that a piece of software integrated with another comprises a system that, like the GPL-licensed portion, should provide those four freedoms.

(Note to the usual people complaining about the FSF "forcing" people to pick their licence: the consequence is that the "other" portion of such a system need only be available under a compatible licence, and the author of that work need only care about the matter if they are doing the combining, anyway. Thus, the Solaris people can just say, "No, you're not supposed to do that - stop it!")

That's why someone pointed out that choosing one's own definition or understanding of "derived work" doesn't necessarily get anyone off the hook. It's very possible that someone interpreting the licence could arrive at an interpretation that suggests a more far-reaching definition, unless some explicit exception has been stated by the GPL-licensed work's authors.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 12:51 UTC (Tue) by paulj (subscriber, #341) [Link] (6 responses)

The Solaris/IllumOS kernel does not become derived of the Linux KVM driver, of course. However the resulting combined work of IllumOS + Linux KVM surely is derived of BOTH those works, and so any distribution in that form surely must be licensed by each of the copyright holders.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 13:54 UTC (Tue) by nye (guest, #51576) [Link] (5 responses)

>However the resulting combined work of IllumOS + Linux KVM surely is derived of BOTH those works

Your 'surely' here is not universally agreed upon - both IllumOS and KVM were developed independently of each other and neither requires the other. On that basis it's hard to see why either would be considered a derivative of the other just because they happen to be used together.

I expect your argument is that modifying KVM to make it work with IllumOS *makes* it a combined work derivative of both (that would be my argument in your place, anyway). That seems reasonable, but it *also* seems reasonable to conclude that bundling the two together with some modification necessary for technical reasons of interoperability is insufficient for these two separate works to now be considered one single work derivative of both.

I'm not saying 'there is definitely no problem', but I am saying that the situation isn't as clear as you make out, and a sound argument could be made that there is no single combined derivative work.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 14:17 UTC (Tue) by paulj (subscriber, #341) [Link] (4 responses)

They are, separately, independent works, of course. However the KVM announced here is not the stock Linux KVM, but one ported (with significant effort) to work within Illumos. I thought that was reasonably obvious from the blog post. ;) I.e. it is a derived work of Linux KVM and Illumos.

To quote the blog post: "just getting it to compile in any meaningful form would take significant effort", "KVM is so tightly integrated into the Linux kernel", "hard to know when to pull in the Linux implementation of a particular facility or function versus writing our own or making a call to the illumos equivalent (if any)". The blog post seems to be fairly explicit about the fact that much judgement, work and creativity was needed to solve a number of problems. It's very hard to think software copyright wouldn't apply to the apparently substantial amount of code they say they had to modify, re-write and write.

My understanding is that the bar to a work being derived is pretty low. I.e. small patches with obvious, no-other-way fixes might not be worthy of copyright, but if there's any room for creativity then that implies that the work does fall under copyright. Further, when you have a one body of code dependent on a very *specific* other piece of code, then my understanding is that the dependent code derives from the other. The Illumos KVM patch seems, by the blog post, to depend on both the Linux KVM and IllumOS code.

That said, the CDDL and the GPL are both free software licenses, and if all the copyright holders concerned are all happy, then there needn't be a problem.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 18:40 UTC (Tue) by fuhchee (guest, #40059) [Link] (3 responses)

"I thought that was reasonably obvious from the blog post. ;) I.e. it is a derived work of Linux KVM and Illumos."

Why "... and Illumos" ?

SmartOS (based on IllumOS) released - with KVM

Posted Aug 16, 2011 20:47 UTC (Tue) by jone (guest, #62596) [Link] (2 responses)

Perhaps that should really read:

"it is a derived work of Linux, KVM, and Illumos"

realistically KVM (or the KVM modules more specifically) is a derived (and dependent) work from interfaces in the Linux kernel, and the Joyent port is a derived work of certain Linux interfaces (some they chose not to port), KVM interfaces, and the Illumos kernel .. (of course there are derived works in the linux kernel from earlier sun code covered now by the CDDL - but who's counting?)

there's a good talk that bmc did (is doing?) for linux-kvm 2011 forum this week that goes into much more detail on what was involved .. slides here:
http://www.linux-kvm.org/wiki/images/7/71/2011-forum-port...

SmartOS (based on IllumOS) released - with KVM

Posted Aug 17, 2011 15:21 UTC (Wed) by vonbrand (guest, #4458) [Link]

The story told there is that this KVM is a straight derivative from the original code, and thus GPLv2 only (look at the code's notices in Linux). Very bad...

SmartOS (based on IllumOS) released - with KVM

Posted Aug 18, 2011 18:31 UTC (Thu) by hummassa (guest, #307) [Link]

There is no "derived work from interfaces". "Abstraction, Filtration, Comparison" -- Google it, please. If there was such thing, Linux would be a derived work from Unix, remember the SCO case? "errno.h" etc...

SmartOS (based on IllumOS) released - with KVM

Posted Aug 27, 2011 3:50 UTC (Sat) by cas (guest, #52554) [Link] (1 responses)

Thought experiment: if somebody writes a kernel driver for Windows and releases it under the GPL does that mean we should 'call [it] out as bullshit'? Same thing.

If Microsoft distributed such a driver with windows, then they would be bringing their entire kernel under the terms of the GPL. For some strange reason, this is unlikely to ever happen.

If a hardware manufacturer did that, they'd be breaking the terms of the driver's GPL....unless they were the author, or got the author to give them permission to do that - e.g. "this driver code is GPL but you're also allowed to distribute it on new PCs with Windows, as long as you abide by the terms of the GPL as it applies to this driver". Similar to the GPL + KDE special permission we used to see back when KDE had a GPL-incompatible license.

if the driver is distributed by itself for the end-user to install on their own machine, then there's no license problem at all. The GPL does not restrict usage, only redistribution.

SmartOS (based on IllumOS) released - with KVM

Posted Aug 29, 2011 13:27 UTC (Mon) by anselm (subscriber, #2796) [Link]

If Microsoft distributed such a driver with windows, then they would be bringing their entire kernel under the terms of the GPL.

No, they wouldn't. They would be distributing that driver without a valid license, which is a different thing altogether.

Wishful thinking notwithstanding, a piece of software will never automatically be GPLed in its entirety, simply because the manufacturer incorporated some GPL code into it and distributed the result. (Incidentally, this is why the GPL is not a »virus« GPL »infection« doesn't work that way.) The only problem in this case is that the GPL code is not distributed according to its license. This means that (a) the manufacturer may be liable to the author(s) of the GPL code for damages resulting from past unlawful distribution of that code, and (b) the manufacturer needs to do something to avoid further non-compliance with the GPL in the future.

They can of course solve the problem by voluntarily putting their codebase under the GPL, too, but if we're talking »big proprietary package« it is more likely that they will simply remove the offending GPL code and replace it by something with a more tractable license, or make arrangements with the copyright holder to be able to use the code under a different, non-GPL, license.