On Thu, May 04, 2023 at 03:51:16PM -0700, Andrii Nakryiko wrote:
> On Thu, May 4, 2023 at 1:05 PM Alexei Starovoitov
> <alexei.starovoitov@xxxxxxxxx> wrote:
> >
> > On Tue, May 02, 2023 at 04:06:13PM -0700, Andrii Nakryiko wrote:
> > > }
> > >
> > > -static struct bpf_map *array_map_alloc(union bpf_attr *attr)
> > > +static u32 array_index_mask(u32 max_entries)
> > > {
> > > - bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
> > > - int numa_node = bpf_map_attr_numa_node(attr);
> > > - u32 elem_size, index_mask, max_entries;
> > > - bool bypass_spec_v1 = bpf_bypass_spec_v1();
> >
> > static inline bool bpf_bypass_spec_v1(void)
> > {
> > return perfmon_capable();
> > }
> >
> > > + /* unprivileged is OK, but we still record if we had CAP_BPF */
> > > + unpriv = !bpf_capable();
> >
> > map->unpriv flag makes sense as !CAP_BPF,
> > but it's not equivalent to bpf_bypass_spec_v1.
> >
>
> argh, right, it's perfmon_capable() :(
>
> what do you propose? do bpf_capable and perfmon_capable fields for
> each map separately? or keep unpriv and add perfmon_capable
> separately? or any better ideas?..
Instead of map->unpriv I'd add map->bpf_capable and map->perfmon_capable
just like we'll be doing to progs.
