RE: Re: Re: Clarification on bpftool dual licensing

RE: Re: Re: Clarification on bpftool dual licensing

[Martin Kelly]

> On Wed, Nov 17, 2021 at 4:32 PM Martin Kelly
> <martin.kelly@crowdstrike.com> wrote:
> >
> > >
> > > On Tue, Nov 16, 2021 at 2:16 AM Daniel Borkmann
> <daniel@iogearbox.net>
> > > wrote:
> > > >
> > > > On 11/15/21 7:20 PM, Martin Kelly wrote:
> > > > > Hi,
> > > > >
> > > > > I have a question regarding the dual licensing provision of bpftool. I
> > > > > understand that bpftool can be distributed as either GPL 2.0 or BSD 2-
> > > clause.
> > > > > That said, bpftool can also auto-generate BPF code that gets specified
> inline
> > > > > in the skeleton header file, and it's possible that the BPF code generated
> is
> > > > > GPL. What I'm wondering is what happens if bpftool generates GPL-
> licensed
> > > BPF
> > > > > code inside the skeleton header, so that you get a header like this:
> > > > >
> > > > > something.skel.h:
> > > > > /* this file is BSD 2-clause, by nature of dual licensing */
> > > >
> > > > Fwiw, the generated header contains an SPDX identifier:
> > > >
> > > >   /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
> > > >   /* THIS FILE IS AUTOGENERATED! */
> > > >
> > > > > /* THIS FILE IS AUTOGENERATED! */
> > > > >
> > > > > /* standard skeleton definitions */
> > > > >
> > > > > ...
> > > > >
> > > > > s->data_sz = XXX;
> > > > > s->data = (void *)"\
> > > > > <eBPF bytecode, produced by GPL 2.0 sources, specified in binary>
> > > > > ";
> > > > >
> > > > > My guess is that, based on the choice to dual-license bpftool, the header
> is
> > > > > meant to still be BSD 2-clause, and the s->data inline code's GPL license
> is
> > > > > not meant to change the licensing of the header itself, but I wanted to
> > >
> > > Yes, definitely that is the intent (but not a lawyer either).
> >
> >  Thanks everyone, that's what I assumed as well. Any objection to a patch
> clarifying this more explicitly?
> >
> > One other, related question: vmlinux.h (generated by "bpftool btf dump file
> /sys/kernel/btf/vmlinux format c"), does not currently contain a license
> declaration. I assume this would have to be a GPL header, since vmlinux.h
> references many GPL'd Linux kernel structs and similar, though again I'm not a
> lawyer and therefore am not certain. Would you all agree with this? If so, any
> objection to a patch adding an SPDX line to the generated vmlinux.h?
>
> Is vmlinux DWARF data GPL'ed? I certainly hope not. So vmlinux.h
> shouldn't be licensed under GPL.

I have no idea; I had assumed that a struct definition coming from a GPL-licensed header would have to be GPL, but again, not a lawyer, and I could totally be wrong. If not GPL though, what would the license be? Is it just "output of a program" and therefore license-less, even though the output happens to be code?

RE: Re: Re: Clarification on bpftool dual licensing

[Toke Høiland-Jørgensen]

Martin Kelly <martin.kelly@crowdstrike.com> writes:

>> > One other, related question: vmlinux.h (generated by "bpftool btf dump file
>> /sys/kernel/btf/vmlinux format c"), does not currently contain a license
>> declaration. I assume this would have to be a GPL header, since vmlinux.h
>> references many GPL'd Linux kernel structs and similar, though again I'm not a
>> lawyer and therefore am not certain. Would you all agree with this? If so, any
>> objection to a patch adding an SPDX line to the generated vmlinux.h?
>>
>> Is vmlinux DWARF data GPL'ed? I certainly hope not. So vmlinux.h
>> shouldn't be licensed under GPL.
>
> I have no idea; I had assumed that a struct definition coming from a
> GPL-licensed header would have to be GPL, but again, not a lawyer, and
> I could totally be wrong. If not GPL though, what would the license
> be? Is it just "output of a program" and therefore license-less, even
> though the output happens to be code?

Totally not a lawyer either, but:

There's (generally, in many jurisdictions, etc), a minimum bar for when
something is considered a "creative work" and thus copyrightable. Debug
output *could* fall short of this (and thus not be copyrightable at
all). It could also fall under the same "API" umbrella as that famous
Google v Oracle case. Or it could fall under the "syscall exception" of
the kernel source.

I guess it would take a court decision to know either way. IMO it would
make sense if vmlinux.h is not copyrightable for whatever reason, but,
again, IANAL :)

Anyway, while we obviously can't resolve legal matters on the mailing,
we can express the *intention* of the community, which is what the
licensing document is trying to do. So it totally makes sense to mention
vmlinux.h here; the question is what should such a text say?

-Toke