Re: [PATCH v2 bpf-next 1/4] bpf: unprivileged BPF access via /dev/bpf

On Aug 7, 2019, at 2:03 AM, Lorenz Bauer [off-list ref] wrote:

quoted

On Wed, 7 Aug 2019 at 06:24, Andy Lutomirski [off-list ref] wrote:
a) Those that, by design, control privileged operations.  This
includes most attach calls, but it also includes allow_ptr_leaks,
bpf_probe_read(), and quite a few other things.  It also includes all
of the by_id calls, I think, unless some clever modification to the
way they worked would isolate different users' objects.  I think that
persistent objects can do pretty much everything that by_id users
would need, so this isn't a big deal.

Slightly OT, since this is an implementation question: GET_MAP_FD_BY_ID
is useful to iterate a nested map. This isn't covered by rights to
persistent objects,
so it would need some thought.

A call to get an fd to a map referenced by a map to which you already have an fd seems reasonable to me. The new fd would inherit the old fd’s access mode.