Re: [PATCH v2 1/2] seccomp: notify user trap about unused filter

[Kees Cook <keescook@chromium.org>]

On Fri, May 29, 2020 at 01:32:03AM +0200, Jann Horn wrote:
> On Fri, May 29, 2020 at 1:11 AM Kees Cook <keescook@chromium.org> wrote:
> > So, is it safe to detach the filter in release_task()? Has dethreading
> > happened yet? i.e. can we race TSYNC? -- is there a possible
> > inc-from-zero?
> 
> release_task -> __exit_signal -> __unhash_process ->
> list_del_rcu(&p->thread_node) drops us from the thread list under
> siglock, which is the same lock TSYNC uses.

Ah, there it is. I missed the __unhash_process() in __exit_signal, but
once I saw the call to release_task(), I figured it was safe at that
point. So this seems correct:

> > I *think* we can do it
> > before the release_thread() call (instead of after cgroup_release()).

> One other interesting thing that can look at seccomp state is
> task_seccomp() in procfs - that can still happen at this point. At the
> moment, procfs only lets you see the numeric filter state, not the
> actual filter contents, so that's not a problem; but if we ever add a
> procfs interface for dumping seccomp filters (in addition to the
> ptrace interface that already exists), that's something to keep in
> mind.

Right -- but we can just reuse the get/put to pin the filter while
dumping it from proc (there IS someone working on this feature...)

> > (Actually, all our refcount_inc()s should be
> > refcount_inc_not_zero() just for robustness.)
> 
> Eeeh... wouldn't that just make the code more complicated for no good reason?

Sorry, ignore that. I got myself briefly confused -- we're fine;
refcount_inc() already does inc-from-zero checking.

-- 
Kees Cook