[PATCHv2 00/14] xen: fix many long-standing grant mapping bugs

[David Vrabel <david.vrabel@citrix.com>]

This series fixes a number of long-standing bugs in the handling of
grant maps.  Refer to the following for all the details.

  http://xenbits.xen.org/people/dvrabel/grant-improvements-C.pdf

In summary, the important uses that this enables are:

1. Block backends can use networked storage safely.

2. Block backends can use network storage provided by other guests on
   the same host.

3. User space block backends can use direct I/O or asynchronous I/O.

The first two patches are the core MM changes necessary.  I shall be
sending these to the MM maintainers seperately.

Patches #3 and #4 remove existing (broken) mechanisms.  This does
temporarily break some previously working use cases, but it does make
the subsequent additions much easier to review.

As a happy side effect, performance is also likely to be improved in
some areas (but I've not got any measurements yet).  User space
backends using grant mapping should see some good improvements from
reduced overheads and better unmap batching.  VIF to VIF network
traffic may also see a small improvement.

Finally, thanks to Jenny who did much of the implementation.

Changes in v2:
- Add find_page() VMA op instead of pages field (so struct
  vm_area_struct doesn't increase in size.
- Make foreign page tracking arch-independant and improve the API.
- Alloc extra memory (for 32-bit archs) for the (domain, gref) when
  allocating the page (instead of during the map).
- Convert gntdev's lock to a mutex.

David