Hardcore Coding

I've never really had the urge to hack on the Linux kernel (not least because I am the world's worst programmer - Fortran, anyone?) but if I did, I'd certainly be using Greg Kroah-Hartman's Linux Kernel in a Nutshell. To both his and O'Reilly's credit, you can download a copy (cc licence), but obviously buying one would be a good idea, too, for all the obvious reasons.

GPLv3: What Linus, Alan, Greg, Andrew and Dave Said

Few subjects in the world of free software have provoked as much discussion as the new GNU GPLv3 licence. Mostly it's outsiders (like me) sounding off about this, but what do the people involved really think?

I decided to ask them, and write up the result for the Guardian. I was expecting a couple of lines back to my emailed questions if I was lucky, but badly underestimated hacker generosity. Linus and his mates sent me back long and typically thoughtful replies, while RMS wanted to talk about it at length.

Since I was only able to use a tiny fraction of this material in the Guardian article that resulted, I thought it might be a useful contribution to the GPLv3 debate to post it here (with the permission of those concerned).

For length reasons, I've split it up into two postings. Below are the replies of the kernel coders, which I received on 3 October, 2006 (placed in the order in which their authors appear in the recent GPLv3 poll). The interview with RMS can be found above.

Linus Torvalds

I don't think there will necessarily be a lot of _practical_ fallout from it, so in that sense it probably doesn't matter all that much. It's not like we haven't had license "discussions" before (the whole BSD vs GPL flame-war seemed to go on for years back in the early nineties). And in many ways, it's not like the actual split between the "Open Source" and the "Free Software" mentality is in any way new, or even brought about by the GPLv3 license.

So while I think there is still a (admittedly pretty remote) chance of some kind of agreement, I don't think that it's a disaster if we end up with a GPLv2 and a new and incompatible GPLv3. It's not like we haven't had licenses before either, and most of them haven't been compatible.

In some ways, I can even hope that it clears the air for all the stupid tensions to just admit that there are differences of opinion, and that the FSF might even just stop using the name "GNU/Linux", finally admitting that Linux never was a GNU project in the first place.

The real downside, I suspect, is just the confusion by yet another incompatible license - and one that shares the same name (licenses such as OSL and GPL were both open source licenses and they were incompatible with each other, but at least they had clear differentiation in their names).

And there's bound to be some productivity loss from all the inevitable arguments, although in all honesty, it's not like open source developers don't spend a lot of time arguing _anyway_, so maybe that won't be all that big of a factor - just a shift of area rather than any actual new lost time ;)

One of the reasons the thing gets so heated is that people (very much me included) feel very strongly about their licenses. It's more than just a legal paper, it's deeply associated with what people have been working on for in some cases decades. So logically I don't think the disagreement really matters a whole lot, but a lot of it is about being very personally attached to some license choice.

Alan Cox

Ar Maw, 2006-10-03 am 13:57 +0100, ysgrifennodd glyn moody:
> Since it seems likely that the kernel will remain under v2, while the
> rest of GNU goes for v3, I was wondering whether you think this is
> going to cause you and others practical problems in your work on the
> kernel. What about companies and end-users of GNU/Linux: will there

There is no such thing as GNU/Linux. For an article like this it's really important to understand and clarify that (and from the US view also as a trademark matter).

I mean there is no abstract entity even that is properly called "GNU/Linux". It's a bit of spin-doctoring by the FSF to try and link themselves to Linux. Normally its just one of those things they do and people sigh about, but when you look at the licensing debate the distinction is vital. (its also increasingly true that FSF owned code is a minority part of Linux)

Linux is not and never has been an FSF project. I would say the majority of the kernel developers don't buy the FSF political agenda. Linus likewise chose the license for the pragmatic reason it was a good license for the OS, not because he supported the GNU manifesto.

Thus this isn't about the Linux people splitting from the FSF, its a separate project that happens to have been consulted as to whether it would like to use a new, allegedly better, variant of the license it chose.

Linux does use FSF tools but that doesn't make it a GNU project any more than this article will be an IBM project because it was typed on a PC, or a BT project because it used an ADSL line.

The Linux kernel being GPLv2 isn't a problem we can see for the future. It is a distinct work to the applications that run on it, just as Windows kernel is to Windows applications. The more awkward corner cases will be LGPL and similar licenses where you want the benefits and flexibility. The FSF have indicated they understand that and will ensure it works out. The licenses are about having barriers to abuse, not barriers to use.

> be negative consequences for them, or do you think that life will
> just go on as before?

I'm not sure what will happen with the rest of the GPL licensed software world. It really is too early to say because the license is a draft at this point and various areas around patents and optional clauses remain open to correction and improvement.

Most GPL licensed code is not controlled by the FSF and probably has too many contributers to relicense. Stuff that is new or has a few owners might change license if the new license is good. However given that most of the work on the FSF owned projects is done by non FSF people then if the license is bad I imagine all the developers will continue the GPLv2 branch and the FSF will be left out in the cold. The FSF know this too and that's why it takes time to build a new license and consensus.

It may well be the new license is mostly used with new code.

> What's the main problem you have with GPLv3?

For the the kernel there are a few, the big one that is hard to fix is the DRM clause. Right now the GPLv2 covers things like DRM keys in generic language and it means the law can interpret that sanely. Its vague but flexible, which lawyers don't like of course. There isn't any caselaw but out of court settlements support the fact this is enforcable.

The GPLv3 variant is much stronger and it appears to cover things like keys to rented devices where the DRM logic is less clear.

The big one though for the kernel is not a legal matter or even a specifically GPLv3 matter. Many people contributed to the kernel under a set of understood terms. Not only would all those people have to agree to a change in those terms but those terms changing would prevent some of the existing users from continuing to use it in the manner they do now.

You can't really make an agreement like that and then change the rules on people who've contributed time, money and code to the Linux project. I support Linus' assertion that legal issues aside he doesn't have the moral right to change the rules this way.

Greg Kroah-Hartman

> My question concerns the timing of the recent white paper: why was it
> released now, and not at the beginning of the GPLv3 consultation process
> when it might have been able to influence things?

The process is not over, and we still hope to influence things. We would not have written that letter otherwise. The main reason it was not done earlier is that we just did not think it was going to be a problem, as the kernel was not going to change licenses. But the more that we realized this was going to have a problem outside of just the kernel, and affect the whole community, we felt that we should at least voice our opinions.

Also, please note that the DRM issues have changed over time from being very broad (which was at least admirable), to being explicitly targeted at only the Linux kernel. Now the license is worded to try to stop the "tivoization" issue.

This is the where a bootloader or bios determines if the crypto signature of the kernel is acceptable or not before it decides to run it or not. This means that only "approved" kernels that come from the company will run properly on the hardware.

Now this kind of restriction pretty much _only_ affects the kernel, not any other type of program. This is because only if you can control the kernel can you ensure that the system is "secure".

So it seems that the FSF is only targeting the Tivo issue, which us kernel developers have explicitly stated in public that it is acceptable to use _our_ code in this manner. So they are now trying to tell another group (us) what we should do to our code.

As the FSF has no contribution in the Linux kernel, and has nothing to do with it in general, we kernel developers are now a bit upset that someone else is trying to tell us that something we explicitly stated was acceptable use of our code, is suddenly bad and wrong.

> Given that the FSF is unlikely to throw away all the work it has done, or
> even modify it substantially/substantively, do you have any thoughts on
> what's going to happen?

I really have no idea, but I would hope that things change for the better. We are already hearing rumors from the people on the different GPLv3 committees that our statement has had an affect, but we will not know for sure until the next draft comes out.

Andrew Morton

Well gee. We're programmers and we spend our time programming, not swanning around at meetings talking about legal matters and playing politics. We find things like licensing to be rather a distraction, and dull. So most people largely ignored it all.

It was only later in the process when the thing started to take shape, when we saw where it was headed and when we began to hear the concerns of various affected parties that there was sufficient motivation to get involved.

In fact this points at a broad problem with the existing process: I'm sure that a large majority of the people who actually write this code haven't made their opinions felt to the FSF. Yet the FSF presumes to speak for them, and proposes to use their work as ammunition in the FSF's campaigns.

And why haven't these programmers made their opinions known? Some are busy. Many work for overlawyered companies and are afraid that they might be seen to be speaking for their companies. Some don't speak English very well. Almost all of them find it to be rather dull and a distraction.

Dave Miller

For the kernel I'm pretty sure things will go on as they have before.

The problems are most likely for the projects under the GNU Project umbrella. All the copyrights to those projects, such as GCC, Binutils, etc. are all assigned to the GNU Project. So the FSF could, and almost certainly will, make all of those projects use the GPL v3.

As an aside, I will note that originally the FSF used to say that they wanted copyright assigned to them "to make it easier to enforce the GPL in court for software projects under the GNU Project umbrella." But as is clear today, it's also a power thing, in that having all the copyrights assigned to them allows the FSF to choose the licensing of the code as they see fit since they are the copyright holder of the complete work.

At the point of a relicense to GPL v3 for these GNU Project source trees one of two things could happen. Either the developers are OK with this, even if to simply "grin and bear it" and things go on under GPL v3. Or, the developers are unhappy with this, and fork off a GPL v2 copy of the tree and do development there.

In the end, even though they've assigned their copyrights to the FSF, the developers do control the ultimate licensing of these GNU projects. If they don't like GPL v3 and work on the GPL v2 fork instead, the FSF is at a loss because while they can mandate whatever they like such mandates are useless if the developers don't want to contribute to the GPL v3 variant.

So being the ones who do the development work is actually a kind of power which permeates through all of the politics. If the political folks do something stupid, the developers can just take their talent and efforts elsewhere.

I'm more than familiar with this process, since I was part of the group that forked the GCC compiler project many years ago because the majority of the GCC developers found the head maintainer (Richard Kenner) impossible to work with. Although he was quite upset about it, there wasn't much that Richard Stallman and the FSF could do about it. In the end the fork became the "real GCC" under GNU Project umbrella once more.

So the opinion of the developers matters a lot, especially when it comes to licensing. It could get messy is a lot of these projects fork, but the GPL v3 isn't a done deal yet so the FSF still has time to fix things up and make it more palatable to people.

> Alone among those polled for their views on the v2 and v3 you choose
>
> 0 I don't really care at all
>
> why is this when everybody else seems to hold such extreme views on the
> subject? Do you think they're getting worked up over nothing?

First, I think the poll was pretty useless.

The poll asked what people think of the GPL v3 draft, which is by definition in draft state and therefore not ready for final consumption. Of course the GPL v3 still needs some fixing. So asking about actually using it in a major software project right now is totally pointless. What would have been more interesting would have been to ask what the developers think about the "core issues" rather than the specific implementation of those issues in the current GPL v3 draft.

For example, polling on what the kernel developers thought about "keying of the Linux kernel" in the way that Tivo does would have been much more interesting. In my opinion, I believe you would have seen about an even split down the middle on this one. But even the people who are against keying think that the DRM language in the GPL v3 meant to combat this is not done correctly.

Several kernel developers believe that GPL v2 already has enough language to make restrictions such as keying be not allowed.

Personally, I'm against keying and I'm quite unhappy with what Tivo did with the Linux kernel. This kind of keying sets a very bad precedent. For example, in the future vendors could get away with some questionable things using keying. Say a vendor sells a piece of hardware, and provides the GPL source to the kernel drivers of the new things in that piece of hardware. Then, they only allow kernel binaries signed with a special key to load. This makes the publishing of their drivers effectively useless. The vendor still controls everything and nobody gains from the code they've submitted. Nobody can recompile a kernel with their changes and actually test it on their hardware, since they have no way to produce a signed kernel that the device will actually allow to boot. So the value of this "contribution" is absolutely zero. This is, in my opinion, totally against the spirit of the GPL v2.

I would have been perfectly fine with Tivo using another OS for their product. All the world does not have to be Linux, and if Linux's license doesn't suit someone, they are free to not use it.

In most examples I've ever been shown where this kind of lockdown is supposedly "legitimate", the owner of the device is effectively making this lockdown decision. For example I'm OK with electronic voting machines used in real elections having their software locked down. The government owns those machines, and is well within their rights to lock down that hardware in order to ensure a proper and fair election to the public by preventing software tampering.

But if I purchase an electronic voting machine of my own, and it uses the Linux kernel, I very much want to tinker with it, build my own kernels, and try to poke holes in the device. How else could we validate that electronic voting machines are safe if the public has no way to test these claims out for themselves, in particular when such devices use open technologies such as the Linux kernel?