On Sun, Sep 29, 2024 at 9:32 AM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Sun, Sep 29, 2024 at 2:31 AM Eric Long via B4 Relay
> <devnull+i.hack3r.moe@xxxxxxxxxx> wrote:
> >
> > From: Eric Long <i@xxxxxxxxxx>
> >
> > FUNCs do not have sizes, thus currently btf__resolve_size will fail
> > with -EINVAL. Add conditions so that we only update size when the BTF
> > object is not function or function prototype.
> >
> > Signed-off-by: Eric Long <i@xxxxxxxxxx>
> > ---
> > tools/lib/bpf/linker.c | 23 +++++++++++++----------
> > 1 file changed, 13 insertions(+), 10 deletions(-)
> >
> > diff --git a/tools/lib/bpf/linker.c b/tools/lib/bpf/linker.c
> > index 81dbbdd79a7c65a4b048b85e1dba99cb5f7cb56b..cffb388fa40ef054c2661b8363120f8a4d3c3784 100644
> > --- a/tools/lib/bpf/linker.c
> > +++ b/tools/lib/bpf/linker.c
> > @@ -2452,17 +2452,20 @@ static int linker_append_btf(struct bpf_linker *linker, struct src_obj *obj)
> > __s64 sz;
> >
> > dst_var = &dst_sec->sec_vars[glob_sym->var_idx];
> > - /* Because underlying BTF type might have
> > - * changed, so might its size have changed, so
> > - * re-calculate and update it in sec_var.
> > - */
> > - sz = btf__resolve_size(linker->btf, glob_sym->underlying_btf_id);
> > - if (sz < 0) {
> > - pr_warn("global '%s': failed to resolve size of underlying type: %d\n",
> > - name, (int)sz);
> > - return -EINVAL;
> > + t = btf__type_by_id(linker->btf, glob_sym->underlying_btf_id);
> > + if (btf_kind(t) != BTF_KIND_FUNC && btf_kind(t) != BTF_KIND_FUNC_PROTO) {
> > + /* Because underlying BTF type might have
> > + * changed, so might its size have changed, so
> > + * re-calculate and update it in sec_var.
> > + */
> > + sz = btf__resolve_size(linker->btf, glob_sym->underlying_btf_id);
> > + if (sz < 0) {
> > + pr_warn("global '%s': failed to resolve size of underlying type: %d\n",
> > + name, (int)sz);
> > + return -EINVAL;
> > + }
> > + dst_var->size = sz;
>
> Looks like a hack to me.
>
> In the test you're using:
> void *bpf_cast_to_kern_ctx(void *obj) __ksym;
>
> but __weak is missing.
> Is that the reason you're hitting this issue?
This is the real issue, unfortunately. And the reason we have this
size update is due to weak *variables*. When the weak variable is
overridden by a "strong" variable or extern is resolved to a concrete
variable, its underlying type might change (e.g., FWD to STRUCT, but
probably there are other cases). Details are a bit hazy by now.
But in any case, this size updating logic should only apply to
variables, so filtering out FUNCs is the straightforward solution. It
can just be coded a bit more nicely, which I suggested in a separate
reply.
