Re: [PATCH v4 bpf-next 2/2] selftests/bpf: Add is_kernel parameter to LSM/bpf test programs

[Paul Moore]

On Tue, Mar 4, 2025 at 8:26 PM Blaise Boscaccy
<bboscaccy@xxxxxxxxxxxxxxxxxxx> wrote:
> Paul Moore <paul@xxxxxxxxxxxxxx> writes:
> > On Tue, Mar 4, 2025 at 3:31 PM Blaise Boscaccy
> > <bboscaccy@xxxxxxxxxxxxxxxxxxx> wrote:
> >>
> >> The security_bpf LSM hook now contains a boolean parameter specifying
> >> whether an invocation of the bpf syscall originated from within the
> >> kernel. Here, we update the function signature of relevant test
> >> programs to include that new parameter.
> >>
> >> Signed-off-by: Blaise Boscaccy bboscaccy@xxxxxxxxxxxxxxxxxxx
> >> ---
> >>  tools/testing/selftests/bpf/progs/rcu_read_lock.c           | 3 ++-
> >>  tools/testing/selftests/bpf/progs/test_cgroup1_hierarchy.c  | 4 ++--
> >>  tools/testing/selftests/bpf/progs/test_kfunc_dynptr_param.c | 6 +++---
> >>  tools/testing/selftests/bpf/progs/test_lookup_key.c         | 2 +-
> >>  tools/testing/selftests/bpf/progs/test_ptr_untrusted.c      | 2 +-
> >>  tools/testing/selftests/bpf/progs/test_task_under_cgroup.c  | 2 +-
> >>  tools/testing/selftests/bpf/progs/test_verify_pkcs7_sig.c   | 2 +-
> >>  7 files changed, 11 insertions(+), 10 deletions(-)
> >
> > I see that Song requested that the changes in this patch be split out
> > back in the v3 revision, will that cause git bisect issues if patch
> > 1/2 is applied but patch 2/2 is not, or is there some BPF magic that
> > ensures that the selftests will still run properly?
> >
>
> So there isn't any type checking in the bpf program's function
> arguments against the LSM hook definitions, so it shouldn't cause any
> build issues. To the best of my knowledge, the new is_kernel boolean
> flag will end up living in r3. None of the current tests reference
> that parameter, so if we bisected and ended up on the previous commit,
> the bpf test programs would in a worst-case scenario simply clobber that
> register, which shouldn't effect any test outcomes unless a test program
> was somehow dependent on an uninitialized value in a scratch register.

Esh.  With that in mind, I'd argue that the two patches really should
just be one patch as you did before.  The patches are both pretty
small and obviously related so it really shouldn't be an issue.

However, since we need this patchset in order to properly implement
BPF signature verification I'm not going to make a fuss if Song feels
strongly that the selftest changes should be split into their own
patch.

-- 
paul-moore.com