On Thu, Dec 3, 2020 at 5:54 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Thu, Dec 03, 2020 at 12:46:30PM -0800, Andrii Nakryiko wrote:
> > + if (attr->attach_prog_fd) {
> > + dst_prog = bpf_prog_get(attr->attach_prog_fd);
> > + if (IS_ERR(dst_prog)) {
> > + dst_prog = NULL;
> > + attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
> > + if (IS_ERR(attach_btf))
> > + return -EINVAL;
> > + if (!btf_is_kernel(attach_btf)) {
> > + btf_put(attach_btf);
> > + return -EINVAL;
>
> Applied, but please consider follow up with different err code here.
> I think we might support this case in the future.
> Specifying prog's BTF as a base and attach_btf_id within it might make
> user space simpler in some cases. prog's btf covers the whole elf file.
The problem is that there is no link from BTF to bpf_prog. And
multiple instances of bpf_progs can re-use the same BTF object and the
same BTF type ID. That would need to be resolved somehow.
But keeping our options open is a good idea either way. So I'll send a
patch to switch this to -EOPNOTSUPP (I think that's the one we need to
use for user-space).
> Where prog_fd is a specific prog. That narrow scope isn't really necessary.
> So may be return ENOTSUPP here for now? With a hint that this might
> work in the future?
