musl build fixes

  batctl-2014.4.0 fails to build with musl, mostly because of somewhat
weird netinet/*.h headers musl provides, but the problem turns out
to be easy to fix.

musl does not allow including netinet/* and linux/* headers together.
batctl includes netinet/if_ether.h indirectly via net/ethernet.h,
so netinet/if_ether.h must be used instead of linux/if_ether.h.

__be16 and __be32 are linux-specific typedefs for uint16_t and
uint32_t with __attribute__((bitwise)) that has no effect
outside of the kernel. Replacing them with uint16_t and uint32_t
removes dependency on linux/types.h.

Signed-off-by: Alex Suykov <alex.suykov@gmail.com>
  

On Wednesday 01 April 2015 20:18:21 Alex Suykov wrote:
> batctl-2014.4.0 fails to build with musl, mostly because of somewhat
> weird netinet/*.h headers musl provides, but the problem turns out
> to be easy to fix.
> 
> musl does not allow including netinet/* and linux/* headers together.
> batctl includes netinet/if_ether.h indirectly via net/ethernet.h,
> so netinet/if_ether.h must be used instead of linux/if_ether.h.
> 
> __be16 and __be32 are linux-specific typedefs for uint16_t and
> uint32_t with __attribute__((bitwise)) that has no effect
> outside of the kernel. Replacing them with uint16_t and uint32_t
> removes dependency on linux/types.h.
> 
> Signed-off-by: Alex Suykov <alex.suykov@gmail.com>

Nacked-by: Sven Eckelmann <sven@narfation.org>

The __be32, __be16 are from the kernel and used there to check if data was 
correctly converted from host byte order to big endian (and the other way 
around). batctl just uses the packet.h from the kernel module. Removing them 
from the userspace would mean that it gets removed from the kernel module. 
This would mean that someone has send this "weird" commit to the next linux 
kernel maintainer. And I am quite happy that I will not be this person.

See https://lists.open-mesh.org/pipermail/b.a.t.m.a.n/2015-March/012926.html
for a proposal how to provide headers which are not yet part of linux-libc-dev 
(linux kernel uapi headers).

Kind regards,
	Sven
  

Wed, Apr 01, 2015 at 10:02:14PM +0200, Sven Eckelmann wrote:

> The linux/if_ether.h -> netinet/if_ether.h seems to be understandable (but 
> rather unfortunate). 
> 
> I am  not sure why why you include linux/types.h to the different source 
> files. Most likely because this include should actually be in packet.h as 
> mentioned in
>  https://lists.open-mesh.org/pipermail/b.a.t.m.a.n/2015-March/012930.html
>  https://lists.open-mesh.org/pipermail/b.a.t.m.a.n/2015-March/012942.html

I assume any changes in packet.h must be done in batman-adv first, then
ported to batctl, and preferably not the other way around.
And since I was only concerned with batctl, I saw packet.h as read-only.

Moving linux/types.h to packet.h does make sense I think.
However, it also brings linux/if_ether.h there, which immediately
makes musl support difficult. Maybe even impractical.


I'm not saying that's wrong, because I'm not trying to say batctl needs
musl support in the first place. Just another point to consider.
  

On Thursday 02 April 2015 00:53:02 Alex Suykov wrote:
> Wed, Apr 01, 2015 at 10:02:14PM +0200, Sven Eckelmann wrote:
> > The linux/if_ether.h -> netinet/if_ether.h seems to be understandable (but
> > rather unfortunate).
> > 
> > I am  not sure why why you include linux/types.h to the different source
> > files. Most likely because this include should actually be in packet.h as
> > mentioned in
> > 
> >  https://lists.open-mesh.org/pipermail/b.a.t.m.a.n/2015-March/012930.html
> >  https://lists.open-mesh.org/pipermail/b.a.t.m.a.n/2015-March/012942.html
> 
> I assume any changes in packet.h must be done in batman-adv first, then
> ported to batctl, and preferably not the other way around.
> And since I was only concerned with batctl, I saw packet.h as read-only.

Yes, this is the preferable way to modify packet.h. I was asking to understand
if these are needed for something else which I may have missed.

Maybe you can remove the type.h stuff and ask to get it queued in after my
patches are in. At least if this would work

> Moving linux/types.h to packet.h does make sense I think.
> However, it also brings linux/if_ether.h there, which immediately
> makes musl support difficult. Maybe even impractical.

The maintainers rejected the use of __KERNEL__ checks when Markus Pargmann
wanted to use them [1]. So I may have to modify my patch to move bitops and
if_ether.h back to main.h. Maybe you can try to modify your packet.h as seen
in the new version of "[PATCHv3 2/4] batman-adv: Add required to includes to
all files" which I will send in some minutes. Would be nice when this version
+ your linux/if_ether.h -> netinet/if_ether.h would work with musl.

> I'm not saying that's wrong, because I'm not trying to say batctl needs
> musl support in the first place. Just another point to consider.

We can at least try to get support in. But when we have two problems to solve
then it may takes some more rounds and extra discussion :)

Kind regards,
	Sven

[1] https://lists.open-mesh.org/pipermail/b.a.t.m.a.n/2014-December/012699.html
  

Thu, Apr 02, 2015 at 08:39:32AM +0200, Sven Eckelmann wrote:

> > Moving linux/types.h to packet.h does make sense I think.
> > However, it also brings linux/if_ether.h there, which immediately
> > makes musl support difficult. Maybe even impractical.
> 
> The maintainers rejected the use of __KERNEL__ checks when Markus Pargmann
> wanted to use them [1]. So I may have to modify my patch to move bitops and
> if_ether.h back to main.h. Maybe you can try to modify your packet.h as seen
> in the new version of "[PATCHv3 2/4] batman-adv: Add required to includes to
> all files" which I will send in some minutes. Would be nice when this version
> + your linux/if_ether.h -> netinet/if_ether.h would work with musl.

Not only it works, it also turns out to be shortest musl-fixes patch so far.
With linux/types.h in packet.h, I only need to replace linux/if_ether.h
with netinet/if_ether.h to get a successful musl build.