Licensing issues

[Andrew Cooper <Andrew.Cooper3@xxxxxxxxxx>]

I've been cross-checking licensing details, and we have some problems.

1) We install xen/include/public/COPYING into /usr/include/xen/COPYING,
which is not common practice.  The installed file is mostly useless
because it discusses files based on their xen.git paths.

2) We actually use the MIT header for the public interface, but we don't
actually call it by it's common name anywhere.

3) The following files are missing the MIT header:

/usr/include/xen/foreign/x86_32.h
/usr/include/xen/foreign/x86_64.h
/usr/include/xen/foreign/arm32.h
/usr/include/xen/foreign/arm64.h
/usr/include/xen/sys/gntalloc.h
/usr/include/xen/device_tree_defs.h
/usr/include/xen/errno.h

Foreign are autogenerated from headers with MIT licences, so that's an
easy fix.

errno.h was an oversight when we added it.  There's no problem fixing
it, as it is covered by multiple statements elsewhere in the tree.

device_tree_defs.h is problematic.  It came into existence in c/s
1c898a9fec7e4 when some LGPL code was moved out of libxl, and some GPL
code was copied out of Xen.  So there's currently an GPL+LGPL vs MIT
licensing violation.

I have not looked through history, but it's likely that the copyright is
covered by individuals/companies who are still active members of Xen,
and I don't anticipate any problem getting it formally relicensed (it's
just a few constants), but this does need fixing.

sys/gntalloc.h is more complicated.  It's stated as public domain which
is fine for our purposes, but inconsistent with everything else, and we
need to adjust the various files we've got which state that the full
public API is MIT.

But it does raise a different bug.  Why do we have random linux headers
committed in the tree, used by some userspace libraries?

Another error I found while poking around is io/libxenvchan.h.  This is
the interface for the dom->dom vchan shared header so ought to live in
the Xen public headers, but it includes <sys/types.h> for no need,
excluding its use in non-posix usecases.  Also also ought to be fixed
from having a zero length array to a flexible member.


4) I'm going to start putting together some fixes some of this.  Henry:
Relicensing device_tree_defs.h is a release blocker.

I'm also intending to start using SDPX identifiers to save on all the
boilerplate.  They're already used elsewhere.

I'm tempted to put together a check which can run in CI to try and
prevent mistakes like this slipping in again, but it's going to be of
little use until it is used in pre-push testing.

~Andrew