Re: GPLv3 Position Statement

From: Linus Torvalds
Date: Wed Sep 27 2006 - 13:29:19 EST

[ This is not so much really a reply to Alan, as a rant on some of the
issues that Alan takes up ]

On Wed, 27 Sep 2006, Alan Cox wrote:

> Ar Mer, 2006-09-27 am 10:58 +0200, ysgrifennodd Jan Engelhardt:
> > I think Linus once said that he does not consider the kernel to
> > become part of a combined work when an application uses the kernel.
> COPYING top of the kernel source tree.

Yes. But also somethign much more fundamental:

Copyright Law - regardless of country

You can claim anything you damn well want in a license, but in the end,
it's all about "derived works". If something is not a derived work, it
doesn't matter _what_ ownership rights you have, it's simply not an issue.

So even if the kernel had a big neon sign that said "You're bound by this
license in user space too!" (or, more likely, didn't have a sign at all,
like was the case originally), that has absolutely _zero_ legal meaning
for a copyright license. A license cannot cover something just because it
"says so".

For example, I could write a copyright license that said

"This license forbids you from ever copying the kernel, or any
work made by Leo Tolstoy, which is just a pseudonym for the
easter bunny"

and the fact that the license claims to control the works of Leo "the
easter bunny" Tolstoy, claiming so simply doesn't make it so.

And yes, the above is obviously ridiculous, but the point is, it's no more
ridiculous than a license that would claim that it extends to programs
just because you can run then on Linux.

In fact, it's also no more ridiculous than a license that claims it
extends copyright the other way - to the hardware or platform that you run
a program on. From a legal standpoint, such wording is just totally

[ So the wording at the top of the license is a clarification of fact, not
really any kind of change of the license itself. It actually does have
some legal meaning (it shows "intent"), but most importantly it allows
people to not have to even worry about frivolous lawsuits. Nobody can
sue people for not running GPLv2 user-space through normal system calls,
because the statement of intent makes it clear that a judge would throw
out such a suit immediately.

So I think the important thing here to take away is that "frivolous"
part of the lawsuit. The language doesn't actually _change_ the legal
meaning, but it protects against idiots. And a lot of lawyers worry
about idiots and money-grubbing douchebags *cough*SCO*cough*, and
as such obvious clarifications _can_ be useful. ]

Another real-world example of this mis-understanding is that a lot of
people seem to think that the GPLv2 disallows "linking" with non-GPLv2
code. Almost everybody I ever meet say that, and the FSF has written long
pieces on shared libraries etc.

People don't have a clue!

The GPLv2 never _ever_ mentions "linking" or any other technical measure
at all. Doing so would just be stupid (another problem with the GPLv3,
btw). So people who think that the GPLv2 disallows "linking" with
non-GPLv2 code had better go back and read the license again.

Grep for it, if you want to. The word "link" simply DOES NOT EXIST IN THE

(It does exist at the end of the _file_ that contains the license, but
that's not actually the license at all, it's just the "btw, this is how
you might _use_ the license", and while legally that can be used to show
"intent", it does not actually extend the copyright in the work in any
way, shape, or form).

What the GPLv2 actually talks about is _only_ about "derived work". And
whether linking (dynamically, statically, or by using an army of worker
gnomes that re-arrange the bits to their liking) means something is
"derived or not" is a totally different issue, and is not something you
can actually say one way or the other, because it will depend on the

I'm always surprised by how many people talk abut the GPLv2 (and, quite
frankly, about the GPLv3 draft) without actually seemingly having ever
read the damn thing, or, more likely, ever really understood any legal
stuff what-so-ever, or the difference between the _use_ of a license, and
the license itself.

For example, in the GPLv3 discussions, I've seen more than one person
claim that I've used a special magic version of the GPLv2 that doesn't
have the "v2 or any later" clause. Again, those people don't have a _clue_
about what they are talking about. They feel very free in arguing about
other peoples copyrigted works, and the fact that I'm not a lawyer, but
then they ignore the fact that I actually _do_ know what I'm talking
about, and that they don't have a stinking clue.

> No. The definition of a derivative work is a legal one and not a
> technical one.

Exactly. A lot of people don't understand this, and a lot of people think
that "derivative" means "being close". Linking doesn't make something
derivative per se - the same way _not_ linking doesn't make it not be

Now, it is also indisputable that if you _need_ to "link", it's a damn
good sign that something is _very_likely_ to be derivative, but as Alan
points out, you could do the same thing with RPC over a socket, and the
fact that you did it technically differently really makes no real
difference. So linking per se isn't the issue, and never has been.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at