[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CABi2SkXaBv85JF6gTd1w-f_i700YSj5JoK8z605bzd6gbPjKkw@mail.gmail.com>
Date: Wed, 21 Aug 2024 09:27:12 -0700
From: Jeff Xu <jeffxu@...omium.org>
To: Pedro Falcato <pedro.falcato@...il.com>
Cc: Andrew Morton <akpm@...ux-foundation.org>, "Liam R. Howlett" <Liam.Howlett@...cle.com>,
Vlastimil Babka <vbabka@...e.cz>, Lorenzo Stoakes <lorenzo.stoakes@...cle.com>,
Shuah Khan <shuah@...nel.org>, linux-mm@...ck.org, linux-kernel@...r.kernel.org,
linux-kselftest@...r.kernel.org, oliver.sang@...el.com,
torvalds@...ux-foundation.org, Michael Ellerman <mpe@...erman.id.au>,
Kees Cook <kees@...nel.org>
Subject: Re: [PATCH v3 7/7] selftests/mm: add more mseal traversal tests
On Wed, Aug 21, 2024 at 9:20 AM Pedro Falcato <pedro.falcato@...il.com> wrote:
>
> On Wed, Aug 21, 2024 at 4:56 PM Jeff Xu <jeffxu@...omium.org> wrote:
> >
> > Hi Pedro
> >
> > On Fri, Aug 16, 2024 at 5:18 PM Pedro Falcato <pedro.falcato@...il.com> wrote:
> > >
> > > Add more mseal traversal tests across VMAs, where we could possibly
> > > screw up sealing checks. These test more across-vma traversal for
> > > mprotect, munmap and madvise. Particularly, we test for the case where a
> > > regular VMA is followed by a sealed VMA.
> > >
> > > Signed-off-by: Pedro Falcato <pedro.falcato@...il.com>
> > > ---
> > > tools/testing/selftests/mm/mseal_test.c | 111 +++++++++++++++++++++++++++++++-
> > > 1 file changed, 110 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/tools/testing/selftests/mm/mseal_test.c b/tools/testing/selftests/mm/mseal_test.c
> > > index 259bef4945e9..0d4d40fb0f88 100644
> > > --- a/tools/testing/selftests/mm/mseal_test.c
> > > +++ b/tools/testing/selftests/mm/mseal_test.c
> > > @@ -766,6 +766,42 @@ static void test_seal_mprotect_partial_mprotect(bool seal)
> > > REPORT_TEST_PASS();
> > > }
> > >
> > > +static void test_seal_mprotect_partial_mprotect_tail(bool seal)
> > > +{
> > > + void *ptr;
> > > + unsigned long page_size = getpagesize();
> > > + unsigned long size = 2 * page_size;
> > > + int ret;
> > > + int prot;
> > > +
> > > + /*
> > > + * Check if a partial mseal (that results in two vmas) works correctly.
> > > + * It might mprotect the first, but it'll never touch the second (msealed) vma.
> > > + */
> > > +
> > > + setup_single_address(size, &ptr);
> > > + FAIL_TEST_IF_FALSE(ptr != (void *)-1);
> > > +
> > > + if (seal) {
> > > + ret = sys_mseal(ptr + page_size, size);
> > you are allocating 2 pages , and I assume you are sealing the second
> > page, so the size should be page_size.
> > ret = sys_mseal(ptr + page_size, page_size);
>
> Yes, good catch, it appears to be harmless but ofc down to straight luck.
> I'll send a fixup for this and the other mistake down there.
>
> >
> > > + FAIL_TEST_IF_FALSE(!ret);
> > > + }
> > > +
> > > + ret = sys_mprotect(ptr, size, PROT_EXEC);
> > > + if (seal)
> > > + FAIL_TEST_IF_FALSE(ret < 0);
> > > + else
> > > + FAIL_TEST_IF_FALSE(!ret);
> > > +
> > > + if (seal) {
> > > + FAIL_TEST_IF_FALSE(get_vma_size(ptr + page_size, &prot) > 0);
> > > + FAIL_TEST_IF_FALSE(prot == 0x4);
> > To test partial mprotect, the test needs to add the check for the
> > first page to be changed, Also to avoid the merge, a PROT_NONE page
> > can to be added in front.
>
> No, I'm leaving partial mprotect to be undefined. It doesn't make
> sense to constraint ourselves, since POSIX wording is already loose.
>
> >
> > > + }
> > > +
> > > + REPORT_TEST_PASS();
> > > +}
> > > +
> > > +
> > > static void test_seal_mprotect_two_vma_with_gap(bool seal)
> > > {
> > > void *ptr;
> > > @@ -983,6 +1019,41 @@ static void test_seal_munmap_vma_with_gap(bool seal)
> > > REPORT_TEST_PASS();
> > > }
> > >
> > > +static void test_seal_munmap_partial_across_vmas(bool seal)
> > > +{
> > > + void *ptr;
> > > + unsigned long page_size = getpagesize();
> > > + unsigned long size = 2 * page_size;
> > > + int ret;
> > > + int prot;
> > > +
> > > + /*
> > > + * Check if a partial mseal (that results in two vmas) works correctly.
> > > + * It might unmap the first, but it'll never unmap the second (msealed) vma.
> > > + */
> > > +
> > > + setup_single_address(size, &ptr);
> > > + FAIL_TEST_IF_FALSE(ptr != (void *)-1);
> > > +
> > > + if (seal) {
> > > + ret = sys_mseal(ptr + page_size, size);
> > ret = sys_mseal(ptr + page_size, page_size);
> >
> > > + FAIL_TEST_IF_FALSE(!ret);
> > > + }
> > > +
> > > + ret = sys_munmap(ptr, size);
> > > + if (seal)
> > > + FAIL_TEST_IF_FALSE(ret < 0);
> > > + else
> > > + FAIL_TEST_IF_FALSE(!ret);
> > > +
> > > + if (seal) {
> > > + FAIL_TEST_IF_FALSE(get_vma_size(ptr + page_size, &prot) > 0);
> > > + FAIL_TEST_IF_FALSE(prot == 0x4);
> > To test partial unmap, the test needs to add the check for the first
> > page to be freed, Also to avoid the merge, a PROT_NONE page needs to
> > be in front.
>
> I'm not testing partial unmap. Partial unmap does not happen. I have
> told you this before.
>
ok. Then this test should be as below ? (need to add PROT_NONE page
before and after)
size = get_vma_size(ptr, &prot);
FAIL_TEST_IF_FALSE(size == 2 * page_size);
FAIL_TEST_IF_FALSE(prot==0x4)
> >
> > The test_seal_munmap_partial_across_vmas shows the behavior
> > difference with in-loop approach and out-loop approach. Previously,
> > both VMAs will not be freed, now the first VMA will be freed, and the
> > second VMA (sealed) won't.
> >
> > This brings to the line you previously mentioned: [1] and I quote:
> > "munmap is atomic and always has been. It's required by POSIX."
>
> This is still true, the comment was a copy-and-paste mindslip. Please
> read the email thread. It has been fixed up by Andrew.
>
Which thread/patch by Andrew ? Could you please send it to me ? (I
might missed that)
> --
> Pedro
Powered by blists - more mailing lists