Re: [PATCH 5/5] mm/slub, kunit: Add testcase for krealloc redzone and zeroing

From: Feng Tang
Date: Tue Sep 10 2024 - 10:08:44 EST


On Tue, Sep 10, 2024 at 03:29:21PM +0200, Vlastimil Babka wrote:
> On 9/9/24 03:29, Feng Tang wrote:
> > Danilo Krummrich raised issue about krealloc+GFP_ZERO [1], and Vlastimil
> > suggested to add some test case which can sanity test the kmalloc-redzone
> > and zeroing by utilizing the kmalloc's 'orig_size' debug feature.
> >
> > It covers the grow and shrink case of krealloc() re-using current kmalloc
> > object, and the case of re-allocating a new bigger object.
> >
> > User can add "slub_debug" kernel cmdline parameter to test it.
> >
> > [1]. https://lore.kernel.org/lkml/20240812223707.32049-1-dakr@xxxxxxxxxx/
> >
> > Suggested-by: Vlastimil Babka <vbabka@xxxxxxx>
> > Signed-off-by: Feng Tang <feng.tang@xxxxxxxxx>
> > ---
> > lib/slub_kunit.c | 46 ++++++++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 46 insertions(+)
> >
> > diff --git a/lib/slub_kunit.c b/lib/slub_kunit.c
> > index 6e3a1e5a7142..03e0089149ad 100644
> > --- a/lib/slub_kunit.c
> > +++ b/lib/slub_kunit.c
> > @@ -186,6 +186,51 @@ static void test_leak_destroy(struct kunit *test)
> > KUNIT_EXPECT_EQ(test, 1, slab_errors);
> > }
> >
> > +static void test_krealloc_redzone_zeroing(struct kunit *test)
> > +{
> > + char *p;
> > + int i;
> > +
> > + KUNIT_TEST_REQUIRES(test, __slub_debug_enabled());
>
> AFAICS this is insufficient, because the static key may be enabled due to
> debugging enabled for different caches than kmalloc, or it might not include
> both red zone and object tracking.

You are right, that concerned me too. In first version, I make it depend
on CONFIG_SLUB_DEBUG_ON==y, but most user' and distribution's kernel
won't enable it, and user have to rebuild kernel to test. So I changed
to this check finally.

If there is a way to judge whether 'slub_debug' is enabled, that would
solve this issue.

>
> But it should be possible to instead create a fake kmalloc cache of size 64
> and use __kmalloc_cache_noprof() like test_kmalloc_redzone_access()?

Yep, I thought about that, and the problem was the krealloc a new 128B
object.

> > +
> > + /* Allocate a 64B kmalloc object */
> > + p = kzalloc(48, GFP_KERNEL);
> > + if (unlikely(is_kfence_address(p))) {
> > + kfree(p);
> > + return;
> > + }
> > + memset(p, 0xff, 48);
> > +
> > + kasan_disable_current();
> > + OPTIMIZER_HIDE_VAR(p);
> > +
> > + /* Test shrink */
> > + p = krealloc(p, 40, GFP_KERNEL | __GFP_ZERO);
> > + for (i = 40; i < 64; i++)
> > + KUNIT_EXPECT_EQ(test, p[i], SLUB_RED_ACTIVE);
> > +
> > + /* Test grow within the same 64B kmalloc object */
> > + p = krealloc(p, 56, GFP_KERNEL | __GFP_ZERO);
> > + for (i = 40; i < 56; i++)
> > + KUNIT_EXPECT_EQ(test, p[i], 0);
> > + for (i = 56; i < 64; i++)
> > + KUNIT_EXPECT_EQ(test, p[i], SLUB_RED_ACTIVE);
> > +
> > + /* Test grow with allocating a bigger 128B object */
> > + p = krealloc(p, 112, GFP_KERNEL | __GFP_ZERO);
>
> The only downside is that krealloc() here might use kmalloc-128 cache that's
> not doing red zoning and object tracking....

Yes.

> > + if (unlikely(is_kfence_address(p)))
> > + goto exit;
> > +
> > + for (i = 56; i < 112; i++)
> > + KUNIT_EXPECT_EQ(test, p[i], 0);
>
> ... but this test is still valid and necessary
>
> > + for (i = 112; i < 128; i++)
> > + KUNIT_EXPECT_EQ(test, p[i], SLUB_RED_ACTIVE);
>
> ... we might skip this test as the red zoning is not done by __do_krealloc()
> anyway in the alloc_new case.
>
> > +
> > +exit:
> > + kfree(p);
>
> Ideally we'd also validate the fake kmalloc cache we created and expect zero
> slab_errors.
>
> Hopefully this approach works and I'm not missing something...

Yep, this should work. As redzone was tested in earlier check, and
not necessary to be checked again here. Will do some test on this.

Thanks,
Feng