Re: [PATCH 3/6] ebpf: add a way to dump an eBPF program

[Tycho Andersen <tycho.andersen@canonical.com>]

On Wed, Sep 09, 2015 at 05:44:06PM -0700, Andy Lutomirski wrote:
> On Wed, Sep 9, 2015 at 5:13 PM, Tycho Andersen
> <tycho.andersen@canonical.com> wrote:
> > On Wed, Sep 09, 2015 at 04:44:24PM -0700, Andy Lutomirski wrote:
> >> On Wed, Sep 9, 2015 at 3:34 PM, Tycho Andersen
> >> <tycho.andersen@canonical.com> wrote:
> >> >
> >> > Here's a thought,
> >> >
> >> > The set I'm currently proposing effectively separates the ref-counting
> >> > of the struct seccomp_filter from the struct bpf_prog (by necessity,
> >> > since we're referring to filters from fds). What if we went a little
> >> > futher, and made a copy of each seccomp_filter on fork(), keeping it
> >> > pointed at the same bpf_prog but adding some metadata about how it was
> >> > inherited (tsk->seccomp.filter->inheritence_count++ perhaps). This
> >> > would still require this change:
> >>
> >> Won't that break the tsync mechanism?
> >
> > We'll need the change I posted (is_ancestor comparing the underlying
> > bpf_prog instead of the seccomp_filter), but then I think it'll work.
> > I guess we'll need to do some more bookkeeping when we install filters
> > via TSYNC since each thread would need its own seccomp_filter, and
> > we'd also have to decide whether a filter installed via TSYNC was
> > inherited or not.
> >
> > Am I missing something?
> 
> Yes.  I don't think that:
> 
> int fd = [create an ebpf fd];
> if (fork()) {
>   /* Process A */
>   seccomp(attach fd);
>   ...
> } else {
>   /* Process B */
>   seccomp(attach fd);
>   ...
> }
> 
> should result in processes A and B being considered to have the same
> seccomp_filter state.  In particular, I eventually want to make the
> seccomp_filter state be considerably more interesting than just the
> bpf program.
> 
> There's another severe problem, I think.  Suppose that ebpf1 and ebpf2
> are ebpf fds.  If processes C and D start out with no filters at all,
> C attaches ebpf1 and ebpf2, and D attaches just ebpf2, then C and D
> are definitely *not* in the same state, and neither is an ancestor of
> the other.

Ah, yes.

> IOW I really do think that seccomp_filter should have identity.

What if we kept a pointer to the seccomp_filter that was inherited on
fork()? Everything "below" that in the tree is not inherited, and
everything above is. Unfortunately, it's not obvious how to restore
this state.

Tycho