Re: [PATCH bpf-next v9 1/4] bpf: Relax tracing prog recursive attach rules

[Song Liu <song@xxxxxxxxxx>]

On Sun, Dec 17, 2023 at 11:22 AM Jiri Olsa <olsajiri@xxxxxxxxx> wrote:
>
> On Sat, Dec 16, 2023 at 05:31:21PM -0800, Song Liu wrote:
> > On Fri, Dec 15, 2023 at 12:11 PM Dmitrii Dolgov <9erthalion6@xxxxxxxxx> wrote:
> > [...]
> > > diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> > > index eb447b0a9423..e7393674ab94 100644
> > > --- a/include/linux/bpf.h
> > > +++ b/include/linux/bpf.h
> > > @@ -1414,6 +1414,7 @@ struct bpf_prog_aux {
> > >         bool dev_bound; /* Program is bound to the netdev. */
> > >         bool offload_requested; /* Program is bound and offloaded to the netdev. */
> > >         bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */
> > > +       bool attach_tracing_prog; /* true if tracing another tracing program */
> > >         bool func_proto_unreliable;
> > >         bool sleepable;
> > >         bool tail_call_reachable;
> > > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > > index 5e43ddd1b83f..bcc5d5ab0870 100644
> > > --- a/kernel/bpf/syscall.c
> > > +++ b/kernel/bpf/syscall.c
> > > @@ -3040,8 +3040,10 @@ static void bpf_tracing_link_release(struct bpf_link *link)
> > >         bpf_trampoline_put(tr_link->trampoline);
> > >
> > >         /* tgt_prog is NULL if target is a kernel function */
> > > -       if (tr_link->tgt_prog)
> > > +       if (tr_link->tgt_prog) {
> > >                 bpf_prog_put(tr_link->tgt_prog);
> > > +               link->prog->aux->attach_tracing_prog = false;
> > > +       }
> > >  }
> > >
> > >  static void bpf_tracing_link_dealloc(struct bpf_link *link)
> > > @@ -3243,6 +3245,12 @@ static int bpf_tracing_prog_attach(struct bpf_prog *prog,
> > >                 goto out_unlock;
> > >         }
> > >
> > > +       /* Bookkeeping for managing the prog attachment chain */
> > > +       if (tgt_prog &&
> > > +           prog->type == BPF_PROG_TYPE_TRACING &&
> > > +           tgt_prog->type == BPF_PROG_TYPE_TRACING)
> > > +               prog->aux->attach_tracing_prog = true;
> > > +
> > >         link->tgt_prog = tgt_prog;
> > >         link->trampoline = tr;
> > >
> > > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > > index 8e7b6072e3f4..f8c15ce8fd05 100644
> > > --- a/kernel/bpf/verifier.c
> > > +++ b/kernel/bpf/verifier.c
> > > @@ -20077,6 +20077,7 @@ int bpf_check_attach_target(struct bpf_verifier_log *log,
> > >                             struct bpf_attach_target_info *tgt_info)
> > >  {
> > >         bool prog_extension = prog->type == BPF_PROG_TYPE_EXT;
> > > +       bool prog_tracing = prog->type == BPF_PROG_TYPE_TRACING;
> > >         const char prefix[] = "btf_trace_";
> > >         int ret = 0, subprog = -1, i;
> > >         const struct btf_type *t;
> > > @@ -20147,10 +20148,21 @@ int bpf_check_attach_target(struct bpf_verifier_log *log,
> > >                         bpf_log(log, "Can attach to only JITed progs\n");
> > >                         return -EINVAL;
> > >                 }
> > > -               if (tgt_prog->type == prog->type) {
> > > -                       /* Cannot fentry/fexit another fentry/fexit program.
> > > -                        * Cannot attach program extension to another extension.
> > > -                        * It's ok to attach fentry/fexit to extension program.
> > > +               if (prog_tracing) {
> > > +                       if (aux->attach_tracing_prog) {
> > > +                               /*
> > > +                                * Target program is an fentry/fexit which is already attached
> > > +                                * to another tracing program. More levels of nesting
> > > +                                * attachment are not allowed.
> > > +                                */
> > > +                               bpf_log(log, "Cannot nest tracing program attach more than once\n");
> > > +                               return -EINVAL;
> > > +                       }
> >
> > If we add
> >
> > + prog->aux->attach_tracing_prog = true;
> >
> > here. We don't need the changes in syscall.c, right?
> >
> > IOW, we set attach_tracing_prog at program load time, not attach time.
> >
> > Would this work?
>
> I think it'd work.. I can just think of a case where we'd not allow
> to attach fentry program (3) to another fentry (2) even if it's not
> attached, but just loaded, like:
>
>
> load (fentry1 -> kernel function)
>
> load (fentry2 -> fentry1)
>   fentry2->attach_tracing_prog = true
>
> load (fentry3 -> fentry2)
>   if (fentry2->aux->attach_tracing_prog)
>     return -EINVAL
>
>
> I guess it's corner case that does not make much sense, but still it
> feels more natural to me to set it in attach time

If we set attach_tracing_prog at attach time, the following will
succeed:

  load (fentry1 -> kernel function)
  load (fentry2 -> fentry1)
  load (fentry3 -> fentry2)
  attach (fentry1)
  attach (fentry2)
  attach (fentry3)

We can even make attach chain longer, as long as we load
the chain first. This is really confusing to me. So I think we should
set the flag at load() time.

Does this make sense?

Thanks,
Song