Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
MIME-Version: 1.0
References: <20220223231752.52241-1-ppbuk5246@gmail.com> <YhbCGDzlTWp2OJzI@zeniv-ca.linux.org.uk>
 <CAM7-yPTM6FNuT4vs2EuKAKitTWMTHw_XzKVggxQJzn5hqbBHpw@mail.gmail.com>
 <CAM7-yPSk35UoGmRY_rCo2=RryBvwbQEjeWfL2tz1ADUosCXNjw@mail.gmail.com> <878ru1umcu.fsf@email.froward.int.ebiederm.org>
In-Reply-To: <878ru1umcu.fsf@email.froward.int.ebiederm.org>
From:   Yun Levi <ppbuk5246@gmail.com>
Date:   Thu, 24 Feb 2022 09:51:21 +0900
Message-ID: <CAM7-yPTPZXPxhtwvvH6KqpRng2idxZiNCLsJHXbWM4ge1wqsBQ@mail.gmail.com>
Subject: Re: [PATCH] fs/exec.c: Avoid a race in formats
To:     "Eric W. Biederman" <ebiederm@xmission.com>
Cc:     Al Viro <viro@zeniv.linux.org.uk>,
        Kees Cook <keescook@chromium.org>,
        linux-fsdevel@vger.kernel.org,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

> Mostly of what has been happening with binary formats lately is code
> removal.
>
> So I humbly suggest the best defense against misuse by modules is to
> simply remove "EXPORT_SYMBOL(__register_binfmt)".

It could be a solution. but that means the kernel doesn't allow
dynamic binfmt using modules too.
I think the best safe way to remove registered binfmt is ...

unregister binfmt list first ---- (1)
synchronize_rcu_task();
// tasklist stack-check...
unload module.

But for this, there shouldn't happen in the above situation of (1).
If unregister_binfmt has this problem.. I think there is no way to
unload safely for dynamic registered binfmt via module.


On Thu, Feb 24, 2022 at 9:42 AM Eric W. Biederman <ebiederm@xmission.com> wrote:
>
> Yun Levi <ppbuk5246@gmail.com> writes:
>
> > On Thu, Feb 24, 2022 at 8:59 AM Yun Levi <ppbuk5246@gmail.com> wrote:
> >>
> >> On Thu, Feb 24, 2022 at 8:24 AM Al Viro <viro@zeniv.linux.org.uk> wrote:
> >> >
> >> > On Thu, Feb 24, 2022 at 08:17:52AM +0900, Levi Yun wrote:
> >> > > Suppose a module registers its own binfmt (custom) and formats is like:
> >> > >
> >> > > +---------+    +----------+    +---------+
> >> > > | custom  | -> |  format1 | -> | format2 |
> >> > > +---------+    +----------+    +---------+
> >> > >
> >> > > and try to call unregister_binfmt with custom NOT in __exit stage.
> >> >
> >> > Explain, please.  Why would anyone do that?  And how would such
> >> > module decide when it's safe to e.g. dismantle data structures
> >> > used by methods of that binfmt, etc.?
> >> > Could you give more detailed example?
> >>
> >> I think if someone wants to control their own binfmt via "ioctl" not
> >> on time on LOAD.
> >> For example, someone wants to control exec (notification,
> >> allow/disallow and etc..)
> >> and want to enable and disable own's control exec via binfmt reg / unreg
> >> In that situation, While the module is loaded, binfmt is still live
> >> and can be reused by
> >> reg/unreg to enable/disable his exec' control.
> >>
> >> module can decide it's safe to unload by tracing the stack and
> >> confirming whether some tasks in the custom binfmt's function after it
> >> unregisters its own binfmt.
> >>
> >> > Because it looks like papering over an inherently unsafe use of binfmt interfaces..
> >>
> >> I think the above example it's quite a trick and stupid.  it's quite
> >> unsafe to use as you mention.
> >> But, misuse allows that situation to happen without any warning.
> >> As a robustness, I just try to avoid above situation But,
> >> I think it's better to restrict unregister binfmt unregister only when
> >> there is no module usage.
> >
> > And not only stupid exmaple,
> > if someone loadable custom binfmt register in __init and __exit via
> > register and unregister_binfmt,
> > I think that situation could happen.
>
> Mostly of what has been happening with binary formats lately is code
> removal.
>
> So I humbly suggest the best defense against misuse by modules is to
> simply remove "EXPORT_SYMBOL(__register_binfmt)".
>
> Eric