LWN: Comments on "Netgear's open-source router"

GPL violation?

job — Wed, 14 Oct 2009 09:48:10 +0000

The GPL has a stronger chance of enforcing these things since it gives a license to redistribute which you would not have otherwise.

Compare for example a) "I will sell you this software, but you may not use it if you have red hair" and b) "I will sell you this software and give you the right to re-sell it, but you can not re-sell it if you have red hair".

The latter would have a much stronger chance of being valid (I'm thinking fast food chains dress codes for example). I am not a lawyer so can't say for sure of course.

Netgear's open-source router

tajyrink — Sat, 10 Oct 2009 06:38:46 +0000

No, my Broadcom based WRT160N does not have gigabit ports, nor do the specs say anything else. That's one minus in the whole router (besides that it's not the L version).

Netgear's open-source router

eli — Fri, 09 Oct 2009 00:28:51 +0000

I agree, gigabit would be nice, but freedom is worth more than gigabit.

Netgear's open-source router

shemminger — Fri, 09 Oct 2009 00:16:46 +0000

Too bad the Linksys WRT160NL has only 100mbit switch, whereas the Broadcom based non-Linux version WRT160N has gigabit ports.

Re: AFS

dlang — Wed, 07 Oct 2009 18:37:18 +0000

in most cases the firmware image is a filesystem.

I don't believe that it matters if the license text has a 'mere aggregation' clause or not, copyright can only cover copying of the work and derived work. I personally have doubts if clauses like 'you can use this as long as you don't use <unrelated thing>' would stand up, which is the only possible way for the license of one thing to control what else it gets shipped with.

GPL violation?

dlang — Wed, 07 Oct 2009 18:33:16 +0000

in spite of what the music and movie industries are preaching, there are limits to the control that a creator via copyright.

GPL violation?

dwmw2 — Wed, 07 Oct 2009 11:08:02 +0000

"A: Its not up to the GPL to decide things like that. It simply does not have the legal authority. It all depends whether or not the drivers are "derivative works" of the Linux kernel. This is a legal concept/term and is defined by law and court precedent.. its up to a judge to decide what exactly is and is not derivative."

Wrong. The licence can refuse you permission to distribute the GPL'd work itself, with whatever conditions it requires.

Think about this example:

I can write a licence, say for an ebook reader, which says "You may never ship this software with a bundled copy of the Kama Sutra". That licence is perfectly valid. It isn't claiming any copyright on the Kama Sutra. It doesn't require the Kama Sutra to be a derived work of my software. Copyright law says that you may only distribute my program if I permit you to do so, and I'm telling you that you're not allowed to do so if you're shipping it with the Kama Sutra. The copyright I hold on my work allows me to do that. I do have that legal authority.

That's basically what the GPL is doing. It's saying that you can distribute the GPL'd work only if you don't ship it as part of a larger whole which contains non-GPL'd stuff.

It says that if you want to make a collective work, then all parts of that work need to be GPL'd. It's very clear that this applies even when the other parts of that collective work are independent and separate works in themselves, and that the GPL needs to apply to each and every part, regardless of who wrote it.

And just like the Kama Sutra example above, it's not claiming copyright on those other parts, or that they are a derived work of the GPL'd work. It's just that unless you abide by the conditions of the GPL, you do not have permission to distribute the original GPL'd work.

Netgear's open-source router

johill — Wed, 07 Oct 2009 09:55:58 +0000

That 300mpbs is the raw phy data rate, the L2 throughput is nowhere near that, that is typically 80-100mbps.

Re: AFS

edmundo — Wed, 07 Oct 2009 09:39:54 +0000

"most of it was written before linux existed, so that code can't possibly be derived from the kernel"

I think you're misunderstanding how this works. It's irrelevant whether the GPL-incompatible code is derived from the kernel; the relevant question is whether there is a work being distributed that is derived from the kernel and includes the GPL-incompatible code. I think one could put together a non-risible argument that the router firmware is such a work. Of course, one could also argue against that proposition by claiming that the firmware is mere aggregation of separate works.

Enforcement

xoddam — Wed, 07 Oct 2009 08:42:55 +0000

> Has anyone noticed any GPL enforcement for Linux kernel code?

Harald Welte has been doing it for years, the first court injunction was granted in 2004:

http://lmgtfy.com/?q=gpl+enforcement+linux+kernel

GPL violation?

dlang — Wed, 07 Oct 2009 07:14:33 +0000

sharing namespace and address space is defiantly not enough to be a derivative work.

as a trivial example, if it was then every firefox plugin would be a derivative work of firefox.

every DOS application would be a derivative of DOS

there was a time when MAC OS was like this, so every MAC application would be a derivative of the OS (and each other)

derivative work is defined MUCH more narrowly

Enforcement

dlang — Wed, 07 Oct 2009 07:11:38 +0000

lookup gplviolations.org (IIRC) they have been doing gpl enforcement on the kernel for several years very sucessfully.

they have targeted mostly embedded uses where the entire kernel was shipped without the source being available.

so far (in spite of a lot of talk) nobody has made a court case over a pure 'derived work' situation, and I suspect that in large part it's because when talking to lawyers they have found that it's not as clear cut as they would like it to be

Enforcement

pabs — Wed, 07 Oct 2009 06:41:31 +0000

B is a good point; the GPL doesn't mean anything if the copyright holders don't care about enforcing the license. Has anyone noticed any GPL enforcement for Linux kernel code? I know busybox devs and the FSF are active on that front, and I remember grumbling about proprietary drivers but I don't recall any actual GPL enforcement for Linux code.

Netgear's open-source router

ebirdie — Wed, 07 Oct 2009 06:22:41 +0000

The tips about AR* boards, WRT160NL and OpenWRT are great. Thank you!

Just a sidenote, although, I find it a bit odd that many devices with 11n still has one 10/100 port or a 10/100 switch and capability to switch port bonding are in odds. One gigabit port would be better.

BUT it is 802.11n, which sells and is sold, not a decent technical solution.

Or is it that with 802.11n no device can achieve the nominal 300 Mbps rate and the truth is that one 10/100 port is more than capable to handle the traffic from/to wireless?

GPL violation?

simlo — Wed, 07 Oct 2009 06:09:18 +0000

"I will leave you with my reasoning as to why a binary kernel module is a derived work."

Many binary modules is only "derived" in the sense they are compiled with the kernels header files - and this is usually not enough to be deemed a derived work in the legal sense. And therefore NVIDIA can ship closed source video drivers as downloads.

BUT the binary module + kernel IS a derived work. I.e. when they are shipped on the same media such as being shipping as firmware inside a device, GPL covers the whole combination. With the same reasoning, you can't install Linux and a NVIDIA driver on a PC and then give it to somebody: THAT would be a GPL violation.

Then why must not the whole kernel+userspace covered: 1) Because of the explicit exception in the Linux COPYING file. 2) The programs in userspace are clearly decoupled from the kernel through a _standeard_ interface (almost POSIX) and therefore not derived. 3) Even though this has never been tested in court, there is absolutely no legal precedence (settled out of court) to suggest otherwise. And every source involved in GPL tells you that userspace is not covered.

GPL violation?

jmm82 — Wed, 07 Oct 2009 05:09:09 +0000

I guess we could go back and forth all night on this issue as many people often do on lwn, since there is no true answer, yet. Therefore, I will make one last comment. I appreciate your insight in my continued journey towards demystifying the GPL and lawyers for that matter.

The point about system calls was an extreme case, but I feel that given the proper precursory findings on some more boarder line cases(binary lkm) that is the direction we are heading.

I will leave you with my reasoning as to why a binary kernel module is a derived work.

Once a kernel module is linked into the kernel it shares the same namespace and if it dereferences a null pointer we are all F*cked, your code, my code, the the binary code. I guess from a programmer's perspective(me) that is a derived work. If the kernel module was staticly linked in this would probablly not even be a discussion, but they are not.

GPL violation?

dlang — Wed, 07 Oct 2009 04:30:27 +0000

quote:
"actually, there is an explanation that states that system calls are the expected API and so there is no expectation that using them can create a derivative work."

Has there been any court rulings on this yet, I honestly have never looked into this, but it is my personal opinion that if a kernel module *IS* determined in court to be violating GPL, then one could argue the same for a system call. I feel the system call would be less likely to hold up in court.

remember that the issue isn't that it uses the kernel, but instead is it a derivative work of the kernel.

people act as if merely using something makes the new thing a derivative work of the thing being used, but it's not nearly that simple. it's pretty clear that something using a defined interface is almost certainly not a derived work (and system calls are very clearly such an interface)

remember that copyright (which is what we are dealing with here) has some requirements for what can be protected,and it explicitly excludes things that are needed for interoperability (this is why lexmark lost it's case against people making replacement ink cartridges,even though they copied the code bit for bit, lexmark used the entire code binary as a key, so that made it impossible for anything else to be compatible and not use the code), there are requirements for it to be creative (so a list of error definitions in errono.h is not enough for SCO to win a copyright infringement suit against IBM), etc

as programmers (especially in opensource development) we tend to take a very liberal attitude about what is a derivative work, giving credit back to original authors whenever possible (to the extent that they get credit even if not a single line of their code remains), but the courts seem to require much stronger ties

GPL violation?

jmm82 — Wed, 07 Oct 2009 04:11:21 +0000

"actually, there is an explanation that states that system calls are the expected API and so there is no expectation that using them can create a derivative work."

"Linus has stated repeatedly that he doesn't think that such a thing is necessary, and that it makes no change to what is permitted under the plain GPLv2, but it makes people feel more comfortable."

I agree, although it does imply the copyright holders stance on the subject. My initial comment was naive, I will give you that.

"Also, Linus is not supportive of closed modules, although he has stated that it is possible for there to be a closed module that is not a derivative of the kernel."

I never said he did support closed modules, but he has chosen to turn the other cheek.

I agree with everything you have said neither of us will know until this does go to court and when it does it will not matter if the symbol was exported gpl or not, expect for the caveat that the copyright holder must be the one to bring someone to court. This will make for a very interesting situation. Hopefully more vendors will go the direction of madwifi --> ath5k and realize on their own the benefits.

GPL violation?

dlang — Wed, 07 Oct 2009 03:48:19 +0000

actually, there is an explanation that states that system calls are the expected API and so there is no expectation that using them can create a derivative work.

Linus has stated repeatedly that he doesn't think that such a thing is necessary, and that it makes no change to what is permitted under the plain GPLv2, but it makes people feel more comfortable.

Also, Linus is not supportive of closed modules, although he has stated that it is possible for there to be a closed module that is not a derivative of the kernel.

one example is a video driver where the code was written for windows and just has a wrapper around it. the wrapper may be a derivative of the kernel, but the other code can't be as it was written completely independently of the kernel

another example is AFS, most of it was written before linux existed, so that code can't possibly be derived from the kernel

GPL violation?

jmm82 — Wed, 07 Oct 2009 03:26:21 +0000

I half agree with you. Only the copyright holder can enforce the GPL on their code and this is more of a warning that they may intend to do so. Once this goes to court, then we will know which one of us is right on the other half.

If a binary module loses in court against the GPL, the next step is user space programs making systems calls and etc, etc, etc....

Up until now no one has acted and probably because of Linus' stance on the subject, but it really is not Linus' choice. If and when this goes to court I feel the GPL holder will win, but I do not feel this will be a good thing. I strongly support the GPL, but feel that this could scare away a lot of companies.

Netgear's open-source router

eli — Wed, 07 Oct 2009 02:27:00 +0000

Hrm... somehow, part of my comment went missing. It was suppose to say
that, IIRC, the 160NL had twice the flash, RAM, and Mhz as the 54L. Not
sure what happened there.

Netgear's open-source router

eli — Wed, 07 Oct 2009 02:25:00 +0000

That is wonderful news! Now to find time to get mine setup and retire
the 54's :)

GPL violation?

dlang — Wed, 07 Oct 2009 00:41:55 +0000

exporting something gpl_only doesn't make anything using it a derivative of the kernel. it says that the developer who created that interface thinks that it does and so is likely to take action against you if you do use it, but that is all.

GPL violation?

jmm82 — Wed, 07 Oct 2009 00:13:59 +0000

The author of kernel code has the right to export functions as gpl-only and these functions can not be used by proprietary drivers. The base kernel license is *gplv2ish* , but the author can export functions as specified above to clear any ambiguities in interpreting the loading of binary distributed modules and circumvent binary blobs from using their code.

Read link http://www.kernel.org/pub/linux/docs/lkml/#s1-19

GPL violation?

dlang — Tue, 06 Oct 2009 23:57:06 +0000

no it isn't. there is nothing in the kernel license saying that proprietary drivers are allowed.

the most that you will see is developers pointing out that the determination of being a derivative work or not is not as simple as many people want it to be.

GPL violation?

pahanson — Tue, 06 Oct 2009 23:54:44 +0000

I think you're right about the intent of GPL, but the kernel license is more
subtle than this; that is, the module interface is covered under a looser
license to allow you to use proprietary drivers.

Netgear's open-source router

sharms — Tue, 06 Oct 2009 19:41:12 +0000

Where can we find more information, just belkins site or are there other resources?

Netgear's open-source router

nbd — Tue, 06 Oct 2009 19:23:56 +0000

The WRT160NL mentioned below is a decent device, IMHO. There are others that are probably also cheaper, but I don't have a list. You can probably find some of them in the OpenWrt forum.

Netgear's open-source router

deucalion — Tue, 06 Oct 2009 19:14:49 +0000

What models from what brands would you recommend with that chipset that have a decent transmission power?

Any hints would be greatly appreciated... I'm currently stuck with my Asus WL-500g Premium. :-/

Netgear's open-source router

nbd — Tue, 06 Oct 2009 17:51:20 +0000

In latest OpenWrt trunk, we fully support this device, including wifi with 11n in AP mode (using ath9k).

Netgear's open-source router

nbd — Tue, 06 Oct 2009 17:49:51 +0000

I'd recommend getting something based on Atheros AR71xx or AR91xx instead.
That way you'd have some decent 11n gear that you can run software on that is actually open instead of just 'mostly open' ;)

Netgear's open-source router

eli — Tue, 06 Oct 2009 17:46:45 +0000

On the Linksys side, there's the WRT160NL. Interestingly, the "L" model
has completely different internals from the WRT160N, using Atheros instead
of Broadcom, IIRC. There are a number of people interested in getting
support into OpenWRT, etc., but I haven't checked up on the current status
of that.

The WRT160NL is wireless-N, and has a USB port. It also has a serial port
that with a bit of ingenuity can be accessed without opening the case. It
has a JTAG port, but it's .05"-pitch instead of the .1"-pitch I'm used to.
I haven't found headers that will fit it yet (pointers welcome). Spec-
had.

GPL violation?

drag — Tue, 06 Oct 2009 16:09:31 +0000

A: Its not up to the GPL to decide things like that. It simply does not
have
the legal authority. It all depends whether or not the drivers are
"derivative works" of the Linux kernel. This is a legal concept/term and is
defined by law and court precedent.. its up to a judge to decide what
exactly
is and is not derivative.

If the kernel + driver shipping together on a router creates a derivative
work then, yes, it must be GPL. If not then it does not. I don't know if it
is or not.

B: This is the standard way Broadcom and the hundreds of different variants
have worked since the original release of the WRT54g many years ago. Nobody
has seemed to make a fuss about it so far.

I don't like it and I won't by another Broadcom-based router until it is
fixed if I can help it. Been burned in the past because of this and its
just plain irritating.. it does not matter if it is legal or not at this
point to me. (now to the copyright holders it may be another story)

Netgear's open-source router

tzafrir — Tue, 06 Oct 2009 15:36:19 +0000

Why 2.6.21? Anything there that doesn't work with 2.6.31?

GPL violation?

simlo — Tue, 06 Oct 2009 15:31:36 +0000

As I read the GPL you can't distribute a non-GPL driver within a device alongside the kernel. The whole work, i.e. kernel+driver, has to be GPL.

Otherwise it would be legal to use a GPL'ed library and distribute library + a non-GPL program using the library dynamically linked.

Netgear's open-source router

sandbagger — Tue, 06 Oct 2009 15:22:50 +0000

Instead of using Netgear's 2.4-based router, I've recently bought the Belkin 8235 version 2. Its marketted as the "N+ Router" and also has a USB port and mostly open firmware. The best part is that it comes with 2.6.21. Belkin also posts the cross-compiler and source/bin packages up at their site here:
http://www.belkin.com/support/opensource/

So far its been great for hacking away .... gotta love having USB to store a larger filesystem and swap on it ;-)

Netgear's open-source router

Baylink — Tue, 06 Oct 2009 14:38:21 +0000

Aw... they used the *Broadcom* again? Sheeze...

Netgear's open-source router

drag — Tue, 06 Oct 2009 14:30:43 +0000

The big problem here is really the proprietary broadcom wifi drivers. This
means that your stuck using 2.4 kernel.

The combination of proprietary broadcom drivers with the 2.4 kernel has
proven to me to be a disaster in terms of reliability and performance in
the past. This is why on wireless access points they need to be
occasionally rebooted and have latency issues with large amounts of
connections and whatnot.

The reverse engineered b43 drivers are much nicer, but in practice they had
some problems and are removed from the last OpenWRT release last time I
checked (a few weeks ago)

---------------

My response to this is "UGH, they have a screwed up definition of Open
Source". They really need to get the hell away from Broadcom if they want
to actually make something that is OSS friendly. That company just does not
have a clue yet. Its a nice move and it shows that there is customer demand
for open routers and whatnot, but its just a step.

The Marvel ARM platform with Atheros would seem to be a much better
performing and OSS friendly platform approach.

Netgear's open-source router

Baylink — Tue, 06 Oct 2009 14:19:11 +0000

<bucket contents="water" temperature="cold">
There may, though, be a fly in this ointment.

I did a quick lazy google search last night, after seeing the slashdot piece and shipping a note off to Jon -- lazy in that I left off the L -- and my inference is that the unit's not new, and that the non-L model is not all that well reviewed.

If this *is* merely an updated model of the WNR3500, then it might prove worthwhile to do some due diligence on that model's repair incidence and how much things have changed in the interim, if any.
</bucket>