On Thu, Mar 14, 2024 at 2:41 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Thu, Mar 14, 2024 at 1:06 PM Andrii Nakryiko
> <andrii.nakryiko@xxxxxxxxx> wrote:
> > >
> > > > What could work and what I am proposing is to pass a list of bound
> > > > maps in prog_load attributes. Then such maps can be used during the
> > > > verification. For normal maps
> > > >
> > > > prog = prog_load(attr={.bound_maps=maps})
> > > >
> > > > will be semantically the same as
> > > >
> > > > prog = prog_load()
> > > > bpf_prog_bind_map(prog, maps)
> > >
> > > Instead of a whole new api, let's make libbpf insert
> > > ld_imm64 r0, map
> > > as the first insn for this case for now.
> >
> > This sounds like a big hack and unnecessary complication, tbh. I'd
> > like to avoid having to do this in libbpf.
> >
> > But I think we almost have this already supported. In BPF_PROG_LOAD
> > UAPI we have fd_array property, right? I think right now we lazily
> > refcnt referenced maps. But I think it makes total sense to just
> > define that bpf_prog will keep references to all BPF objects passed in
> > through fd_array, WDYT? Verifier will just iterate all provided FDs,
> > determine kind of BPF object it is (and reject unknown ones), and then
> > just take refcnts on each of them once. On prog free we'll just do the
> > same in reverse and be done with it.
> >
> > It also can be used as a batch and single-step (in the sense it will
> > be done as part of program load instead of a separate command)
> > alternative for bpf_prog_bind_map(), I suppose.
>
> fd_array approach also works. There can be map and btf fds in there.
> I would only bind maps this way.
Any reason why we should have non-uniform behavior between maps and
BTFs? Seems more error-prone to have a difference here, tbh.
