On Tue, Jul 04, 2023 at 05:36:26PM +0200, Florent Revest wrote:
> I checked with the original author, the mmap_FIXED test case wasn't
> properly tested and fails. Currently, it maps two consecutive (non
> overlapping) pages and expects the second mapping to be denied by MDWE
> but these two pages have nothing to do with each other so MDWE is
> actually out of the picture here.
>
> What the test actually intended to do was to remap a virtual address
> using MAP_FIXED. However, this operation unmaps the existing mapping and
> creates a new one so the va is backed by a new page and MDWE is again
> out of the picture, all remappings should succeed.
>
> This patch keeps the test case to make it clear that this situation is
> expected to work.
>
> Signed-off-by: Florent Revest <revest@xxxxxxxxxxxx>
> Reviewed-by: David Hildenbrand <david@xxxxxxxxxx>
> Fixes: 4cf1fe34fd18 ("kselftest: vm: add tests for memory-deny-write-execute")
> ---
> tools/testing/selftests/mm/mdwe_test.c | 9 +++------
> 1 file changed, 3 insertions(+), 6 deletions(-)
>
> diff --git a/tools/testing/selftests/mm/mdwe_test.c b/tools/testing/selftests/mm/mdwe_test.c
> index d0954c657feb..91aa9c3099e7 100644
> --- a/tools/testing/selftests/mm/mdwe_test.c
> +++ b/tools/testing/selftests/mm/mdwe_test.c
> @@ -168,13 +168,10 @@ TEST_F(mdwe, mmap_FIXED)
> self->p = mmap(NULL, self->size, PROT_READ, self->flags, 0, 0);
> ASSERT_NE(self->p, MAP_FAILED);
>
> - p = mmap(self->p + self->size, self->size, PROT_READ | PROT_EXEC,
> + /* MAP_FIXED unmaps the existing page before mapping which is allowed */
> + p = mmap(self->p, self->size, PROT_READ | PROT_EXEC,
> self->flags | MAP_FIXED, 0, 0);
> - if (variant->enabled) {
> - EXPECT_EQ(p, MAP_FAILED);
> - } else {
> - EXPECT_EQ(p, self->p);
> - }
> + EXPECT_EQ(p, self->p);
> }
This is just validating the MDWE doesn't block a MAP_FIXED replacement?
Reviewed-by: Kees Cook <keescook@xxxxxxxxxxxx>
-Kees
>
> TEST_F(mdwe, arm64_BTI)
> --
> 2.41.0.255.g8b1d071c50-goog
>
--
Kees Cook
