Re: [PATCH bpf-next v9 02/10] bpf: Return false for bpf_prog_check_recur() default case

[Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx>]

On Mon, Nov 4, 2024 at 10:02 PM Yonghong Song <yonghong.song@xxxxxxxxx> wrote:
>
> >
> > I also don't understand the point of this patch 2.
> > The patch 3 can still do:
> >
> > + switch (prog->type) {
> > + case BPF_PROG_TYPE_KPROBE:
> > + case BPF_PROG_TYPE_TRACEPOINT:
> > + case BPF_PROG_TYPE_PERF_EVENT:
> > + case BPF_PROG_TYPE_RAW_TRACEPOINT:
> > +   return PRIV_STACK_ADAPTIVE;
> > + default:
> > +   break;
> > + }
> > +
> > + if (!bpf_prog_check_recur(prog))
> > +   return NO_PRIV_STACK;
> >
> > which would mean that iter, lsm, struct_ops will not be allowed
> > to use priv stack.
>
> One example is e.g. a TC prog. Since bpf_prog_check_recur(prog)
> will return true (means supporting recursion), and private stack
> does not really support TC prog, the logic will become more
> complicated.
>
> I am totally okay with removing patch 2 and go back to my
> previous approach to explicitly list prog types supporting
> private stack.

The point of reusing bpf_prog_check_recur() is that we don't
need to duplicate the logic.
We can still do something like:
switch (prog->type) {
 case BPF_PROG_TYPE_KPROBE:
 case BPF_PROG_TYPE_TRACEPOINT:
 case BPF_PROG_TYPE_PERF_EVENT:
 case BPF_PROG_TYPE_RAW_TRACEPOINT:
    return PRIV_STACK_ADAPTIVE;
 case BPF_PROG_TYPE_TRACING:
 case BPF_PROG_TYPE_LSM:
 case BPF_PROG_TYPE_STRUCT_OPS:
    if (bpf_prog_check_recur())
      return PRIV_STACK_ADAPTIVE;
    /* fallthrough */
  default:
    return NO_PRIV_STACK;
}