Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 10 Feb 2022 13:09:13 -0800
From:   Kees Cook <keescook@chromium.org>
To:     Jann Horn <jannh@google.com>
Cc:     "Eric W. Biederman" <ebiederm@xmission.com>,
        Robert =?utf-8?B?xZp3acSZY2tp?= <robert@swiecki.net>,
        stable@vger.kernel.org, Andy Lutomirski <luto@amacapital.net>,
        Will Drewry <wad@chromium.org>, linux-kernel@vger.kernel.org,
        linux-hardening@vger.kernel.org, Oleg Nesterov <oleg@redhat.com>
Subject: Re: [PATCH 1/3] signal: HANDLER_EXIT should clear SIGNAL_UNKILLABLE
Message-ID: <202202101254.1174AB2B@keescook>
References: <20220210025321.787113-1-keescook@chromium.org>
 <20220210025321.787113-2-keescook@chromium.org>
 <CAG48ez1m7XJ1wJvTHtNorH480jTWNgdrn5Q1LTZZQ4uve3r4Sw@mail.gmail.com>
 <202202100935.FB3E60FA5@keescook>
 <CAG48ez3fG7S1dfE2-JAtyOZUK=0_iZ03scf+oD6gwVyD1Qp33g@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <CAG48ez3fG7S1dfE2-JAtyOZUK=0_iZ03scf+oD6gwVyD1Qp33g@mail.gmail.com>
Precedence: bulk

On Thu, Feb 10, 2022 at 07:01:39PM +0100, Jann Horn wrote:
> On Thu, Feb 10, 2022 at 6:37 PM Kees Cook <keescook@chromium.org> wrote:
> > On Thu, Feb 10, 2022 at 05:18:39PM +0100, Jann Horn wrote:
> > > On Thu, Feb 10, 2022 at 3:53 AM Kees Cook <keescook@chromium.org> wrote:
> > > > Fatal SIGSYS signals were not being delivered to pid namespace init
> > > > processes. Make sure the SIGNAL_UNKILLABLE doesn't get set for these
> > > > cases.
> > > >
> > > > Reported-by: Robert Święcki <robert@swiecki.net>
> > > > Suggested-by: "Eric W. Biederman" <ebiederm@xmission.com>
> > > > Fixes: 00b06da29cf9 ("signal: Add SA_IMMUTABLE to ensure forced siganls do not get changed")
> > > > Cc: stable@vger.kernel.org
> > > > Signed-off-by: Kees Cook <keescook@chromium.org>
> > > > ---
> > > >  kernel/signal.c | 5 +++--
> > > >  1 file changed, 3 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/kernel/signal.c b/kernel/signal.c
> > > > index 38602738866e..33e3ee4f3383 100644
> > > > --- a/kernel/signal.c
> > > > +++ b/kernel/signal.c
> > > > @@ -1342,9 +1342,10 @@ force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
> > > >         }
> > > >         /*
> > > >          * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
> > > > -        * debugging to leave init killable.
> > > > +        * debugging to leave init killable, unless it is intended to exit.
> > > >          */
> > > > -       if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
> > > > +       if (action->sa.sa_handler == SIG_DFL &&
> > > > +           (!t->ptrace || (handler == HANDLER_EXIT)))
> > > >                 t->signal->flags &= ~SIGNAL_UNKILLABLE;
> > >
> > > You're changing the subclause:
> > >
> > > !t->ptrace
> > >
> > > to:
> > >
> > > (!t->ptrace || (handler == HANDLER_EXIT))
> > >
> > > which means that the change only affects cases where the process has a
> > > ptracer, right? That's not the scenario the commit message is talking
> > > about...
> >
> > Sorry, yes, I was not as accurate as I should have been in the commit
> > log. I have changed it to:
> >
> > Fatal SIGSYS signals (i.e. seccomp RET_KILL_* syscall filter actions)
> > were not being delivered to ptraced pid namespace init processes. Make
> > sure the SIGNAL_UNKILLABLE doesn't get set for these cases.
> 
> So basically force_sig_info() is trying to figure out whether
> get_signal() will later on check for SIGNAL_UNKILLABLE (the SIG_DFL
> case), and if so, it clears the flag from the target's signal_struct
> that marks the process as unkillable?
> 
> This used to be:
> 
> if (action->sa.sa_handler == SIG_DFL)
>     t->signal->flags &= ~SIGNAL_UNKILLABLE;
> 
> Then someone noticed that in the ptrace case, the signal might not
> actually end up being consumed by the target process, and added the
> "&& !t->ptrace" clause in commit
> eb61b5911bdc923875cde99eb25203a0e2b06d43.
> 
> And now Robert Swiecki noticed that that still didn't accurately model
> what'll happen in get_signal().
> 
> This seems hacky to me, and also racy: What if, while you're going
> through a SECCOMP_RET_KILL_PROCESS in an unkillable process, some
> other thread e.g. concurrently changes the disposition of SIGSYS from
> a custom handler to SIG_DFL?

Do you mean after force_sig_info_to_task() has finished but before
get_signal()? SA_IMMUTABLE will block changes to the action.

If you mean before force_sig_info_to_task(), I don't see how that's
possible since it's under lock:

        if (blocked || ignored || (handler != HANDLER_CURRENT)) {
                action->sa.sa_handler = SIG_DFL;
                if (handler == HANDLER_EXIT)
                        action->sa.sa_flags |= SA_IMMUTABLE;
	...
        if (action->sa.sa_handler == SIG_DFL &&
            (!t->ptrace || (handler == HANDLER_EXIT)))
                t->signal->flags &= ~SIGNAL_UNKILLABLE;

Given handler = HANDLER_EXIT, it'll always be SIG_DFL.

> Instead of trying to figure out whether the signal would have been
> fatal without SIGNAL_UNKILLABLE, I think it would be better to find a
> way to tell the signal-handling code that SIGNAL_UNKILLABLE should be
> bypassed for this specific signal, or something along those lines...
> but of course that's also kind of messy because the signal-sending
> code might fall back to just using the pending signal mask on
> allocation failure IIRC?

My original patch aimed that way:

diff --git a/kernel/signal.c b/kernel/signal.c
index 9b04631acde8..c124a09de6de 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2787,7 +2787,8 @@ bool get_signal(struct ksignal *ksig)
 		 * case, the signal cannot be dropped.
 		 */
 		if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
-				!sig_kernel_only(signr))
+				!sig_kernel_only(signr) &&
+				!(ka->sa.sa_flags & SA_IMMUTABLE))
 			continue;
 
 		if (sig_kernel_stop(signr)) {

But I don't think there's a race, and Eric's suggestion seemed
better in the sense that the state change is entirely contained by
force_sig_info_to_task().

-Kees

-- 
Kees Cook