Re: [PATCH v2 bpf-next 01/16] bpf: Introduce bpf_sys_bpf() helper and program type.

[Alexei Starovoitov <alexei.starovoitov@gmail.com>]

On Fri, Apr 23, 2021 at 11:16 AM Yonghong Song <yhs@fb.com> wrote:
> > +
> > +static bool syscall_prog_is_valid_access(int off, int size,
> > +                                      enum bpf_access_type type,
> > +                                      const struct bpf_prog *prog,
> > +                                      struct bpf_insn_access_aux *info)
> > +{
> > +     if (off < 0 || off >= U16_MAX)
> > +             return false;
>
> Is this enough? If I understand correctly, the new program type
> allows any arbitrary context data from user as long as its size
> meets the following constraints:
>     if (ctx_size_in < prog->aux->max_ctx_offset ||
>             ctx_size_in > U16_MAX)
>                 return -EINVAL;
>
> So if user provides a ctx with size say 40 and inside the program looks
> it is still able to read/write to say offset 400.
> Should we be a little more restrictive on this?

At the load time the program can have a read/write at offset 400,
but it will be rejected at prog_test_run time.
That's similar to tp and raw_tp test_run-s and attach-es.
That's why test_run has that check you've quoted.
It's a two step verification.
The verifier rejects <0 || > u16_max right away and
keeps the track of max_ctx_offset.
Then at attach/test_run the final check is done with an actual ctx_size_in.