How much Linux Kernel code is GPL 2 only? At most 60%

Linus and other Linux kernel developers have made it very clear that the kernel will continue to use the GNU GPL version 2, and not move to the GPL 3. But the fact that the kernel is 'effectively' GPL 2 does not mean that all of its code is GPL 2 only - by this I mean that some of the code may be dual-licensed. To get a working Linux system, you must of course use the GPL 2 license. But with dual-licensed code you can use another license when you use it somewhere else, say if you want to use the code in another kernel (perhaps OpenSolaris). But how much of the Linux kernel code is dual-licensed? That is what I tried to find out.

I am not a lawyer. But I do meddle with code. So, to investigate this, I wrote a short program to scan the Linux kernel code (version 2.6.20) and check what licenses appear in it. The code is in Python. If anyone is interested, I am willing to release it. Update: Following requests, the source code is available here.

You may be tempted to skip to the results, but you should probably read the following explanations of what exactly I did first.

Basically, the program I wrote reads the license statements at the beginning of each file, and tries to match them against patterns. This is extremely 'hackish'; I just randomly sampled large amounts of files and kept patching my code until it worked well. Given this method, please note that the results are only an estimate. I did not read each file myself, and while I manually checked a large amount of samples, there is room for some statistical error. However, I do believe that the results are more or less indicative of the true numbers.

I scanned only source files, not headers (hopefully I am not mistaken in doing so). I decided to measure the 'amount' of code under each license by, well, the amount of code - measured in file sizes. So, the total sum of file sizes is what counts, not the number of files or the number of lines in them.

I separated the files into 4 categories:

1. GPL 2 only. These are files that say that only the GPL 2 may be used.
2. GPL 2 or above. These files say explicitly that the GPL 2 may be used, or any later version. I included GPL/BSD dual-licensed code here, since it is (I presume) going to be compatible with the GPL 3.
   
3. GPL, unspecified. These files just say "licensed under the GPL", or such (see examples below), without specifying any version number. There is some debate about what this means. Some believe that such code can use any GPL license (2, 3, etc.), while others believe otherwise, particularly since Linus has a general statement (in the COPYING file), saying that
   

   "Also note that the only valid version of the GPL as far as the kernel is concerned is _this_ particular version of the license (ie v2, not v2.2 or v3.x or whatever), unless explicitly otherwise stated."

   (Interestingly, Linus deliberately makes room for dual-licensing.) Now, as mentioned before, I am not a lawyer, so I do not know the status of this issue. However, when a source file said "GPL, see the file COPYING" (or such), I treated this as "GPL 2 only". So things fall under "unspecified" only when all a file says is "GPL'ed", with no further explanation.
4. Other. Some files didn't mention a license at all, those fall into this category (perhaps they fall under the GPL 2, due to the COPYING file?). Esoteric license phrasings may also fall in here (of which there are few; I worked to get my code working with the vast majority of files).

Ok, after much introduction, here are the numbers for the Linux kernel, version 2.6.20:

GPL 2 only:	32,215,150 bytes
GPL 2 or above:	60,637,907 bytes
GPL, unspecified:	19,773,264 bytes
Other:	43,762,840 bytes

A first remark must acknowledge the total size of all of these numbers: the Linux kernel is enormous (as I guess we would expect for the operating system that supports the most hardware out of the box). Now, as for the various licenses: there seems to be quite a lot of "GPL 2 or above" code, even if we do assume that the 'unspecified' and 'other' sections are "GPL 2 only" - almost 40% of the kernel code is explicitly GPL 2 or above.

So, it seems that a GPL 3 operating system should be able to use a large part of the Linux kernel code. I am of course thinking mainly of OpenSolaris here (which would be interested mainly in device driver code, and not the rest; perhaps later I'll run the program just on the drivers). The implications of OpenSolaris using the GPL 3 were discussed in my previous post, but I didn't have any figures to talk about back then. But now, the picture is somewhat clearer. We may have interesting times ahead.


Notes:

When sifting through the kernel code, I noticed several things that caught my eye. Here are a few. Emphasis is always my own (but how would text files be emphasized, anyhow). I admit that these notes are mostly boring, except for the final one, which is amusing. You may want to skip ahead.

Let's get started. Some licenses are, well, confused. For example, /arch/arm/mach-integrator/integrator_cp.c contains

This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License.

- and that is where the license ends. Either version 2, or what, I asked myself. Perhaps this is the result of some copy&pasting gone wrong. In any case I counted these as "GPL 2 only".

Some files are extremely brief in their licensing. For example, /Documentation/pcmcia/crc32hash.c states

crc32hash.c - derived from linux/lib/crc32.c, GNU GPL v2

Pretty concise. Of course, I counted this as GPL 2 only. Other phrasings are brief but specify no license: /Documentation/networking/ifenslave.c has

This file is under the GPL.

More extreme examples write only "GPL'd". These fell under "GPL, unspecified".

Some files are extremely specific in their licenses, e.g. /drivers/net/atp.c writes

This software may be used and distributed according to the terms of the GNU General Public License (GPL), incorporated herein by reference. Drivers based on or derived from this code fall under the GPL and must retain the authorship, copyright and license notice. This file is not a complete program and may only be used when the entire operating system is licensed under the GPL.

Very detailed - but no GPL version is given. So this is 'GPL, unspecified'.

Worthy of note are other dual-licensing schemes. I saw code that was both BSD and GPL, and code that was both GPL and MIT/XFree86. However, the vast majority is GPL, of one sort or another. Yet, there was one amusing exception: /drivers/md/dm-log.c has the following license:

/*
* Copyright (C) 2003 Sistina Software
*
* This file is released under the LGPL.
*/

LGPL, and not GPL. Are there no legal issues with that? Turns out that there aren't, since LGPL code can be converted to GPL as needed.