[v6,0/5] KUnit-KASAN Integration

Message

David Gow April 18, 2020, 3:18 a.m. UTC

This patchset contains everything needed to integrate KASAN and KUnit.

KUnit will be able to:
(1) Fail tests when an unexpected KASAN error occurs
(2) Pass tests when an expected KASAN error occurs

Convert KASAN tests to KUnit with the exception of copy_user_test
because KUnit is unable to test those.

Add documentation on how to run the KASAN tests with KUnit and what to
expect when running these tests.

Depends on "[PATCH v3 kunit-next 0/2] kunit: extend kunit resources
API" patchset [1]

Changes from v5:
 - Split out the panic_on_warn changes to a separate patch.
 - Fix documentation to fewer to the new Kconfig names.
 - Fix some changes which were in the wrong patch.
 - Rebase on top of kselftest/kunit (currently identical to 5.7-rc1)

Changes from v4:
 - KASAN no longer will panic on errors if both panic_on_warn and
   kasan_multishot are enabled.
 - As a result, the KASAN tests will no-longer disable panic_on_warn.
 - This also means panic_on_warn no-longer needs to be exported.
 - The use of temporary "kasan_data" variables has been cleaned up
   somewhat.
 - A potential refcount/resource leak should multiple KASAN errors
   appear during an assertion was fixed.
 - Some wording changes to the KASAN test Kconfig entries.

Changes from v3:
 - KUNIT_SET_KASAN_DATA and KUNIT_DO_EXPECT_KASAN_FAIL have been
 combined and included in KUNIT_DO_EXPECT_KASAN_FAIL() instead.
 - Reordered logic in kasan_update_kunit_status() in report.c to be
 easier to read.
 - Added comment to not use the name "kasan_data" for any kunit tests
 outside of KUNIT_EXPECT_KASAN_FAIL().

Changes since v2:
 - Due to Alan's changes in [1], KUnit can be built as a module.
 - The name of the tests that could not be run with KUnit has been
 changed to be more generic: test_kasan_module.
 - Documentation on how to run the new KASAN tests and what to expect
 when running them has been added.
 - Some variables and functions are now static.
 - Now save/restore panic_on_warn in a similar way to kasan_multi_shot
 and renamed the init/exit functions to be more generic to accommodate.
 - Due to [2] in kasan_strings, kasan_memchr, and
 kasan_memcmp will fail if CONFIG_AMD_MEM_ENCRYPT is enabled so return
 early and print message explaining this circumstance.
 - Changed preprocessor checks to C checks where applicable.

Changes since v1:
 - Make use of Alan Maguire's suggestion to use his patch that allows
   static resources for integration instead of adding a new attribute to
   the kunit struct
 - All KUNIT_EXPECT_KASAN_FAIL statements are local to each test
 - The definition of KUNIT_EXPECT_KASAN_FAIL is local to the
   test_kasan.c file since it seems this is the only place this will
   be used.
 - Integration relies on KUnit being builtin
 - copy_user_test has been separated into its own file since KUnit
   is unable to test these. This can be run as a module just as before,
   using CONFIG_TEST_KASAN_USER
 - The addition to the current task has been separated into its own
   patch as this is a significant enough change to be on its own.


[1] https://lore.kernel.org/linux-kselftest/[e-mail]/T/#t
[2] https://bugzilla.kernel.org/show_bug.cgi?id=206337

David Gow (1):
  mm: kasan: Do not panic if both panic_on_warn and kasan_multishot set

Patricia Alfonso (4):
  Add KUnit Struct to Current Task
  KUnit: KASAN Integration
  KASAN: Port KASAN Tests to KUnit
  KASAN: Testing Documentation

 Documentation/dev-tools/kasan.rst |  70 +++
 include/kunit/test.h              |   5 +
 include/linux/kasan.h             |   6 +
 include/linux/sched.h             |   4 +
 lib/Kconfig.kasan                 |  18 +-
 lib/Makefile                      |   3 +-
 lib/kunit/test.c                  |  13 +-
 lib/test_kasan.c                  | 682 +++++++++++++-----------------
 lib/test_kasan_module.c           |  76 ++++
 mm/kasan/report.c                 |  37 +-
 10 files changed, 513 insertions(+), 401 deletions(-)
 create mode 100644 lib/test_kasan_module.c

Comments

David Gow April 24, 2020, 6:36 a.m. UTC | #1

On Sat, Apr 18, 2020 at 11:28 PM Alan Maguire <[e-mail]> wrote:
>
>
> On Fri, 17 Apr 2020, David Gow wrote:
>
> > This patchset contains everything needed to integrate KASAN and KUnit.
> >
> > KUnit will be able to:
> > (1) Fail tests when an unexpected KASAN error occurs
> > (2) Pass tests when an expected KASAN error occurs
> >
> > Convert KASAN tests to KUnit with the exception of copy_user_test
> > because KUnit is unable to test those.
> >
>
> I tried building and running and things look good but I am
> still seeing the three failures I reported before, even with
> CONFIG_AMD_MEM_ENCRYPT not set.  My config is attached if you
> want to try and reproduce at your end.  Oddly this config was
> working before IIRC (once CONFIG_AMD_MEM_ENCRYPT was not set).
>
> Here's the failures:
>
>    # kasan_memchr: EXPECTATION FAILED at lib/test_kasan.c:545
>     Expected fail_data.report_expected == fail_data.report_found, but
>         fail_data.report_expected == 1
>         fail_data.report_found == 0
>     not ok 31 - kasan_memchr
>     # kasan_memcmp: EXPECTATION FAILED at lib/test_kasan.c:566
>     Expected fail_data.report_expected == fail_data.report_found, but
>         fail_data.report_expected == 1
>         fail_data.report_found == 0
>     not ok 32 - kasan_memcmp

I was able to reproduce these (along with a kasan_strings) failure,
and the cause seems to be some combination of __builtin functions
being inlined by the compiler and potentially dead code elimination,
as fixed by Daniel Axtens here:
https://lkml.org/lkml/2020/4/23/708

I've sent out v7 of the patchset[1], which I've rebased on top of
Daniel's patches, and can no longer reproduce those test failures with
your .config.

Cheers,
-- David

[1]: https://lkml.org/lkml/2020/4/24/80