On Tue, Apr 4, 2023 at 5:02 PM Andrii Nakryiko
<andrii.nakryiko@xxxxxxxxx> wrote:
>
> On Tue, Apr 4, 2023 at 7:51 AM David Vernet <void@xxxxxxxxxxxxx> wrote:
> >
> > On Mon, Apr 03, 2023 at 09:50:21PM -0700, Alexei Starovoitov wrote:
> > > From: Alexei Starovoitov <ast@xxxxxxxxxx>
> > >
> > > The patch set is addressing a fallout from
> > > commit 6fcd486b3a0a ("bpf: Refactor RCU enforcement in the verifier.")
> > > It was too aggressive with PTR_UNTRUSTED marks.
> > > Patches 1-6 are cleanup and adding verifier smartness to address real
> > > use cases in bpf programs that broke with too aggressive PTR_UNTRUSTED.
> > > The partial revert is done in patch 7 anyway.
> > >
> > > Alexei Starovoitov (8):
> > > bpf: Invoke btf_struct_access() callback only for writes.
> > > bpf: Remove unused arguments from btf_struct_access().
> > > bpf: Refactor btf_nested_type_is_trusted().
> > > bpf: Teach verifier that certain helpers accept NULL pointer.
> > > bpf: Refactor NULL-ness check in check_reg_type().
> > > bpf: Allowlist few fields similar to __rcu tag.
> > > bpf: Undo strict enforcement for walking untagged fields.
> > > selftests/bpf: Add tracing tests for walking skb and req.
> >
> > For whole series:
> >
> > Acked-by: David Vernet <void@xxxxxxxxxxxxx>
>
> Added David's acks manually (we really need to teach pw-apply to do
> this automatically...) and applied.
+1
I was hoping that patchwork will add this feature eventually,
but it seems faster to hack the pw-apply script instead.
> I've added a single sentence to
> patch #1 with why (I think) btf_struct_access() callback
> simplification was done, I didn't want to hold the patch set just due
> to that, as the rest looked good. But please do consider renaming the
> callback to more write-access implying name as a follow up, as current
> situation with the same name but different semantics is confusing.
>
> Applied to bpf-next, thanks.
Thanks.
Renaming either the btf_struct_access() function or (*btf_struct_access) field
was on the todo list as a potential workaround,
since this name caused a weird issue with clang in LTO build.
For some reason two global symbols were generated.
Yonghong is investigating.
fwiw btf_struct_write_access sounds fine as a new name.
