On Wed, 18 Nov 2020 at 02:43, Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote:
>
> On Tue, Nov 17, 2020 at 12:29 AM Björn Töpel <bjorn.topel@xxxxxxxxx> wrote:
> >
> > Some architectures have strict alignment requirements. In that case,
> > the BPF verifier detects if a program has unaligned accesses and
> > rejects them. A user can pass BPF_F_ANY_ALIGNMENT to a program to
> > override this check. That, however, will only work when a privileged
> > user loads a program. A unprivileged user loading a program with this
> > flag will be rejected prior entering the verifier.
>
> I'd include this paragraph as a code comment right next to the check below.
>
> >
> > Hence, it does not make sense to load unprivileged programs without
> > strict alignment when testing the verifier. This patch avoids exactly
> > that.
> >
> > Signed-off-by: Björn Töpel <bjorn.topel@xxxxxxxxx>
> > ---
> > tools/testing/selftests/bpf/test_verifier.c | 12 +++++++++---
> > 1 file changed, 9 insertions(+), 3 deletions(-)
> >
> > diff --git a/tools/testing/selftests/bpf/test_verifier.c b/tools/testing/selftests/bpf/test_verifier.c
> > index 9be395d9dc64..2075f6a98813 100644
> > --- a/tools/testing/selftests/bpf/test_verifier.c
> > +++ b/tools/testing/selftests/bpf/test_verifier.c
> > @@ -1152,9 +1152,15 @@ static void get_unpriv_disabled()
> >
> > static bool test_as_unpriv(struct bpf_test *test)
> > {
> > - return !test->prog_type ||
> > - test->prog_type == BPF_PROG_TYPE_SOCKET_FILTER ||
> > - test->prog_type == BPF_PROG_TYPE_CGROUP_SKB;
> > + bool req_aligned = false;
> > +
> > +#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
> > + req_aligned = test->flags & F_NEEDS_EFFICIENT_UNALIGNED_ACCESS;
> > +#endif
> > + return (!test->prog_type ||
> > + test->prog_type == BPF_PROG_TYPE_SOCKET_FILTER ||
> > + test->prog_type == BPF_PROG_TYPE_CGROUP_SKB) &&
> > + !req_aligned;
>
> It's a bit convoluted. This seems a bit more straightforward:
>
> #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
> if (test->flags & F_NEEDS_EFFICIENT_UNALIGNED_ACCESS)
> return false;
> #endif
> /* the rest of logic untouched */
>
> ?
>
Ugh. Yes, indeed. *blush*
Björn
