Re: [RFC bpf-next 3/5] bpf: pin, translate, and unpin __kptr_user from syscalls.

[Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx>]

On Mon, Aug 12, 2024 at 10:24 AM Kui-Feng Lee <sinquersw@xxxxxxxxx> wrote:
>
>
> >> +static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
> >> +                               void *key, void *value, __u64 flags)
> >> +{
> >> +       int err;
> >> +
> >> +       if (flags & BPF_FROM_USER) {
> >
> > there shouldn't be a need for this extra flag.
> > map->record has the info whether uptr is present or not.
>
> The BPF_FROM_USER flag is used to support updating map values from BPF
> programs as well. Although BPF programs can udpate map values, I
> don't want the values of uptrs to be changed by the BPF programs.
>
> Should we just forbid the BPF programs to udpate the map values having
> uptrs in them?

hmm. map_update_elem() is disallowed from bpf prog.

        case BPF_MAP_TYPE_TASK_STORAGE:
                if (func_id != BPF_FUNC_task_storage_get &&
                    func_id != BPF_FUNC_task_storage_delete &&
                    func_id != BPF_FUNC_kptr_xchg)
                        goto error;