2024-04-02 13:38:54

by Guenter Roeck

Subject: [PATCH] mm/slub, kunit: Use inverted data to corrupt kmem cache

Two failure patterns are seen randomly when running slub_kunit tests with
CONFIG_SLAB_FREELIST_RANDOM and CONFIG_SLAB_FREELIST_HARDENED enabled.

Pattern 1:
# test_clobber_zone: pass:1 fail:0 skip:0 total:1
ok 1 test_clobber_zone
# test_next_pointer: EXPECTATION FAILED at lib/slub_kunit.c:72
Expected 3 == slab_errors, but
slab_errors == 0 (0x0)
# test_next_pointer: EXPECTATION FAILED at lib/slub_kunit.c:84
Expected 2 == slab_errors, but
slab_errors == 0 (0x0)
# test_next_pointer: pass:0 fail:1 skip:0 total:1
not ok 2 test_next_pointer

In this case, test_next_pointer() overwrites p[s->offset], but the data
at p[s->offset] is already 0x12.

Pattern 2:
ok 1 test_clobber_zone
# test_next_pointer: EXPECTATION FAILED at lib/slub_kunit.c:72
Expected 3 == slab_errors, but
slab_errors == 2 (0x2)
# test_next_pointer: pass:0 fail:1 skip:0 total:1
not ok 2 test_next_pointer

In this case, p[s->offset] has a value other than 0x12, but one of the
expected failures is nevertheless missing.

Invert data instead of writing a fixed value to corrupt the cache data
structures to fix the problem.

Fixes: 1f9f78b1b376 ("mm/slub, kunit: add a KUnit test for SLUB debugging functionality")
Cc: Oliver Glitta <[email protected]>
Cc: Vlastimil Babka <[email protected]>
CC: Daniel Latypov <[email protected]>
Cc: Marco Elver <[email protected]>
Signed-off-by: Guenter Roeck <[email protected]>
---
lib/slub_kunit.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/slub_kunit.c b/lib/slub_kunit.c
index d4a3730b08fa..4ce960438806 100644
--- a/lib/slub_kunit.c
+++ b/lib/slub_kunit.c
@@ -55,7 +55,7 @@ static void test_next_pointer(struct kunit *test)

ptr_addr = (unsigned long *)(p + s->offset);
tmp = *ptr_addr;
- p[s->offset] = 0x12;
+ p[s->offset] = ~p[s->offset];

/*
* Expecting three errors.
--
2.39.2

2024-04-04 14:27:44

by Vlastimil Babka

Subject: Re: [PATCH] mm/slub, kunit: Use inverted data to corrupt kmem cache

On 4/2/24 3:38 PM, Guenter Roeck wrote:
> Two failure patterns are seen randomly when running slub_kunit tests with
> CONFIG_SLAB_FREELIST_RANDOM and CONFIG_SLAB_FREELIST_HARDENED enabled.
>
> Pattern 1:
> # test_clobber_zone: pass:1 fail:0 skip:0 total:1
> ok 1 test_clobber_zone
> # test_next_pointer: EXPECTATION FAILED at lib/slub_kunit.c:72
> Expected 3 == slab_errors, but
> slab_errors == 0 (0x0)
> # test_next_pointer: EXPECTATION FAILED at lib/slub_kunit.c:84
> Expected 2 == slab_errors, but
> slab_errors == 0 (0x0)
> # test_next_pointer: pass:0 fail:1 skip:0 total:1
> not ok 2 test_next_pointer
>
> In this case, test_next_pointer() overwrites p[s->offset], but the data
> at p[s->offset] is already 0x12.
>
> Pattern 2:
> ok 1 test_clobber_zone
> # test_next_pointer: EXPECTATION FAILED at lib/slub_kunit.c:72
> Expected 3 == slab_errors, but
> slab_errors == 2 (0x2)
> # test_next_pointer: pass:0 fail:1 skip:0 total:1
> not ok 2 test_next_pointer
>
> In this case, p[s->offset] has a value other than 0x12, but one of the
> expected failures is nevertheless missing.
>
> Invert data instead of writing a fixed value to corrupt the cache data
> structures to fix the problem.
>
> Fixes: 1f9f78b1b376 ("mm/slub, kunit: add a KUnit test for SLUB debugging functionality")
> Cc: Oliver Glitta <[email protected]>
> Cc: Vlastimil Babka <[email protected]>
> CC: Daniel Latypov <[email protected]>
> Cc: Marco Elver <[email protected]>
> Signed-off-by: Guenter Roeck <[email protected]>

Added to slab/for-next, thanks.

> ---
> lib/slub_kunit.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/lib/slub_kunit.c b/lib/slub_kunit.c
> index d4a3730b08fa..4ce960438806 100644
> --- a/lib/slub_kunit.c
> +++ b/lib/slub_kunit.c
> @@ -55,7 +55,7 @@ static void test_next_pointer(struct kunit *test)
>
> ptr_addr = (unsigned long *)(p + s->offset);
> tmp = *ptr_addr;
> - p[s->offset] = 0x12;
> + p[s->offset] = ~p[s->offset];
>
> /*
> * Expecting three errors.