Re: [PATCH v2 01/10] kunit: tool: Enable CONFIG_FORTIFY_SOURCE under UML

[Kees Cook <keescook@chromium.org>]

On Mon, May 22, 2023 at 12:43:51PM -0700, Nick Desaulniers wrote:
> On Wed, May 10, 2023 at 12:24 PM Kees Cook <keescook@chromium.org> wrote:
> >
> > On Fri, Apr 07, 2023 at 04:42:27PM -0700, Nick Desaulniers wrote:
> > > On Fri, Apr 7, 2023 at 4:33 PM Nick Desaulniers <ndesaulniers@google.com> wrote:
> > > >
> > > > On Fri, Apr 7, 2023 at 12:27 PM Kees Cook <keescook@chromium.org> wrote:
> > > > >
> > > > > Since commit ba38961a069b ("um: Enable FORTIFY_SOURCE"), it's possible
> > > > > to run the FORTIFY tests under UML. Enable CONFIG_FORTIFY_SOURCE when
> > > > > running with --altests to gain additional coverage, and by default under
> > > >
> > > > two L's in alltest?
> > >
> > > Also, while testing this series:
> > > ```
> > > $ LLVM=1 ./tools/testing/kunit/kunit.py run
> > > ...
> > > [16:40:09] ================== fortify (24 subtests) ===================
> > > [16:40:09] [PASSED] known_sizes_test
> > > [16:40:09] [PASSED] control_flow_split_test
> > > [16:40:09] [PASSED] alloc_size_kmalloc_const_test
> > > [16:40:09]     # alloc_size_kmalloc_dynamic_test: EXPECTATION FAILED
> > > at lib/fortify_kunit.c:249
> > > [16:40:09]     Expected __builtin_dynamic_object_size(p, 1) == expected, but
> > > [16:40:09]         __builtin_dynamic_object_size(p, 1) == -1
> > > (0xffffffffffffffff)
> > > [16:40:09]         expected == 11 (0xb)
> > > [16:40:09] __alloc_size() not working with __bdos on kmemdup("hello
> > > there", len, gfp)
> >
> > I'm still tracking this down. I'm not sure what's happening here, but it
> > seems to be Clang-specific, and due to some interaction with the changes
> > I made for Kunit examination. WHY it happens I haven't found yet.
> 
> Was this what exposed https://github.com/llvm/llvm-project/issues/62789?

Nope -- I found this while working on:
https://lore.kernel.org/lkml/20230517225838.never.965-kees@kernel.org/

i.e. I was surprised I could use static initializers with a flexible
array, and then I went and verified various related behaviors between
GCC and Clang.

-- 
Kees Cook