Re: [PATCH bpf-next 1/5] bpf: Handle 8-byte values in DEVMAP and DEVMAP_HASH

[Jesper Dangaard Brouer <brouer@redhat.com>]

On Wed, 27 May 2020 08:27:36 -0600
David Ahern <dsahern@gmail.com> wrote:

> On 5/27/20 4:26 AM, Jesper Dangaard Brouer wrote:
> > IMHO we really need to leverage BTF here, as I'm sure we need to do more
> > extensions, and this size matching will get more and more unmaintainable.
> > 
> > With BTF in place, dumping the map via bpftool, will also make the
> > fields "self-documenting".  
> 
> furthermore, the kernel is changing the value - an fd is passed in and
> an id is returned. I do not see how any of this fits into BTF.

It can, as BTF actually support union's (I just tested that).

For the sake of end-users understanding this, I do wonder if it is
better to define the struct without the union, and have longer names
that will be part of BTF description, e.g (dumped via bpftool):

 struct dev_map_ext_val {
        u32 ifindex;
        int bpf_prog_fd_write;
        u32 bpf_prog_id_read;
 };

But a union would also work (also tested via BPF loading and BTF dumpinmg):

 struct dev_map_ext_val {
        u32 ifindex;
        union {
                int bpf_prog_fd_write;
                u32 bpf_prog_id_read;
        };
 };

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer