Re: [PATCH] lib: overflow: Convert to Kunit

From: Kees Cook
Date: Thu Feb 17 2022 - 12:09:36 EST

On Wed, Feb 16, 2022 at 02:57:12PM -0800, Daniel Latypov wrote:
> On Wed, Feb 16, 2022 at 2:42 PM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> >
> > Convert overflow unit tests to KUnit, for better integration into the
> > kernel self test framework. Includes a rename of test_overflow.c to
> > overflow_kunit.c, and CONFIG_TEST_OVERFLOW to CONFIG_OVERFLOW_KUNIT_TEST.
> >
> > $ ./tools/testing/kunit/kunit.py config
> > ...
> > $ ./tools/testing/kunit/kunit.py run overflow
>
> JFYI, you can run this as a one-liner via
>
> $ ./tools/testing/kunit/kunit.py run --kunitconfig /dev/stdin <<EOF
> CONFIG_KUNIT=y
> CONFIG_TEST_OVERFLOW=y
> EOF
>
> The above is taken from my own duplicate version of this patch
> [1] https://lore.kernel.org/linux-kselftest/20210503211536.1384578-1-dlatypov@xxxxxxxxxx/

Ah-ha! I thought I remembered this conversion being proposed before but
I totally failed to find it. Thank you! I'll compare/adjust this patch
and add you as Co-developed-by.

> > ...
> > [14:33:51] Starting KUnit Kernel (1/1)...
> > [14:33:51] ============================================================
> > [14:33:51] ================== overflow (11 subtests) ==================
> > [14:33:51] [PASSED] u8_overflow_test
> > [14:33:51] [PASSED] s8_overflow_test
> > [14:33:51] [PASSED] u16_overflow_test
> > [14:33:51] [PASSED] s16_overflow_test
> > [14:33:51] [PASSED] u32_overflow_test
> > [14:33:51] [PASSED] s32_overflow_test
> > [14:33:51] [PASSED] u64_overflow_test
> > [14:33:51] [PASSED] s64_overflow_test
> > [14:33:51] [PASSED] overflow_shift_test
> > [14:33:51] [PASSED] overflow_allocation_test
> > [14:33:51] [PASSED] overflow_size_helpers_test
> > [14:33:51] ==================== [PASSED] overflow =====================
> > [14:33:51] ============================================================
> > [14:33:51] Testing complete. Passed: 11, Failed: 0, Crashed: 0, Skipped: 0, Errors: 0
> > [14:33:51] Elapsed time: 12.525s total, 0.001s configuring, 12.402s building, 0.101s running
> >
> > Cc: David Gow <davidgow@xxxxxxxxxx>
> > Cc: Vitor Massaru Iha <vitor@xxxxxxxxxxx>
> > Cc: Rasmus Villemoes <linux@xxxxxxxxxxxxxxxxxx>
> > Cc: Nick Desaulniers <ndesaulniers@xxxxxxxxxx>
> > Co-developed-by: Vitor Massaru Iha <vitor@xxxxxxxxxxx>
> > Signed-off-by: Vitor Massaru Iha <vitor@xxxxxxxxxxx>
> > Link: https://lore.kernel.org/lkml/20200720224418.200495-1-vitor@xxxxxxxxxxx/
> > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
>
> Reviewed-by: Daniel Latypov <dlatypov@xxxxxxxxxx>
>
> Looks good to me, some minor nits/suggestions wrt KUnit usage.
> Nice to see this test converted over!

Thanks!

> [...]
> > index f6530fce799d..4cc27b9926a1 100644
> > --- a/lib/test_overflow.c
> > +++ b/lib/overflow_kunit.c
> > @@ -1,9 +1,13 @@
> > // SPDX-License-Identifier: GPL-2.0 OR MIT
> > /*
> > - * Test cases for arithmetic overflow checks.
> > + * Test cases for arithmetic overflow checks. See:
> > + * https://www.kernel.org/doc/html/latest/dev-tools/kunit/kunit-tool.html#configuring-building-and-running-tests
> > + * ./tools/testing/kunit/kunit.py config
> > + * ./tools/testing/kunit/kunit.py run overflow [--raw_output]
> > */
> > #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
>
> We can drop the pr_fmt now, I think

My instinct is to leave these in place just so that anything weird that
gets inlined and sneaks a pr_*() call into the code will have a
meaningful prefix.

> [...]
> > @@ -510,30 +477,28 @@ static int __init test_ ## func (void *arg) \
> > \
> > /* Tiny allocation test. */ \
> > ptr = alloc ## want_arg ## want_gfp ## want_node (func, arg, 1);\
> > - if (!ptr) { \
> > - pr_warn(#func " failed regular allocation?!\n"); \
> > - return 1; \
> > - } \
> > + KUNIT_EXPECT_FALSE_MSG(test, !ptr, \
> > + #func " failed regular allocation?!\n"); \
>
> Optional: we can consider using KUNIT_ASSERT_NOT_ERR_OR_NULL_MSG() here.
> It's a more heavy handed than just a `return` on failure, but if the
> regular allocation failed, we're probably justified in bailing out on
> the whole test case.

Yeah, I think it might work here. Earlier I hadn't figured out how to
convert each test separately, but now an ASSERT makes sense.

>
> > + if (!ptr) \
> > + return; \
> > free ## want_arg (free_func, arg, ptr); \
> > \
> > /* Wrapped allocation test. */ \
> > ptr = alloc ## want_arg ## want_gfp ## want_node (func, arg, \
> > a * b); \
> > - if (!ptr) { \
> > - pr_warn(#func " unexpectedly failed bad wrapping?!\n"); \
> > - return 1; \
> > - } \
> > + KUNIT_EXPECT_FALSE_MSG(test, !ptr, \
> > + #func " unexpectedly failed bad wrapping?!\n"); \
> > + if (!ptr) \
> > + return; \
> > free ## want_arg (free_func, arg, ptr); \
> > \
> > /* Saturated allocation test. */ \
> > ptr = alloc ## want_arg ## want_gfp ## want_node (func, arg, \
> > array_size(a, b)); \
> > - if (ptr) { \
> > - pr_warn(#func " missed saturation!\n"); \
> > + KUNIT_EXPECT_FALSE_MSG(test, ptr, \
> > + #func " missed saturation!\n"); \
> > + if (ptr) \
>
> We can instead do
>
> if (ptr) {
> KUNIT_FAIL(test, #func "missed saturation!");
> free...()
> }
>
> IMO, it's a bit easier to read that way, but not that important.

Ah yes, good. That's much better.

I will respin and resend...

--
Kees Cook