Received-SPF: pass (google.com: domain of linux-bluetooth-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210720104905.6870-1-penguin-kernel@I-love.SAKURA.ne.jp>
 <CABBYNZLt=fJTtdj9qvC22GkF_uYFe59D2bS3u61K14=Gq43Qbw@mail.gmail.com> <bb9ceb4d-9e5c-1487-233b-426bc58e9a91@i-love.sakura.ne.jp>
In-Reply-To: <bb9ceb4d-9e5c-1487-233b-426bc58e9a91@i-love.sakura.ne.jp>
From:   Luiz Augusto von Dentz <luiz.dentz@gmail.com>
Date:   Fri, 23 Jul 2021 14:28:01 -0700
Message-ID: <CABBYNZJnC-b0s5LW=zTLh4_bV44Uv2-6LXXy+vRAgQmxk=defg@mail.gmail.com>
Subject: Re: [PATCH] Bluetooth: reorganize ioctls from hci_sock_bound_ioctl()
To:     Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc:     Marcel Holtmann <marcel@holtmann.org>,
        Johan Hedberg <johan.hedberg@gmail.com>,
        "linux-bluetooth@vger.kernel.org" <linux-bluetooth@vger.kernel.org>,
        LinMa <linma@zju.edu.cn>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Linus Torvalds <torvalds@linux-foundation.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Tetsuo,

On Wed, Jul 21, 2021 at 4:42 PM Tetsuo Handa
<penguin-kernel@i-love.sakura.ne.jp> wrote:
>
> On 2021/07/22 3:17, Luiz Augusto von Dentz wrote:
> > I think it would have been cleaner if we have dedicated functions for
> > each command like I did in my patch:
> >
> > https://patchwork.kernel.org/project/bluetooth/patch/20210717000731.3836303-1-luiz.dentz@gmail.com/
>
> But your patch was proven to be unsafe. There is a use-after-unregister
> race window which would require at least 1000 lines of modification and
> a lot of careful review if we try to manage without my patch.
> Such all-in-one-step change is too much for "sleep in atomic context"
> regression fix which is preventing syzbot from testing Bluetooth module
> and is preventing Linux distributors from fixing CVE-2021-3573.

Im not saying you should adopt my solution, the locking etc stay the
same as in this set but each command should have a helper function to
make it clearer that way we don't have to re-evaluate the command over
and over.

> As far as I can see, it is lock_sock() (not bh_lock_sock_nested() in your
> patch) that is needed for protecting
>
>         sk->sk_err = EPIPE;
>         sk->sk_state = BT_OPEN;
>         sk->sk_state_change(sk);
>
> in hci_sock_dev_event(HCI_DEV_UNREG) from concurrent modification
>
>         lock_sock(sk);
>
>         if (sk->sk_state == BT_BOUND) {
>                 err = -EALREADY;
>                 goto done;
>         }
>
>         (...snipped...)
>
> -               hci_pi(sk)->hdev = hdev;
> +               if (hdev) {
> +                       hci_pi(sk)->dev = hdev->id;
> +                       hci_dev_put(hdev);
> +               }
>
>         (...snipped...)
>         /* Race window is here. */
>         (...snipped...)
>
>         sk->sk_state = BT_BOUND;
> done:
>         release_sock(sk);
>
> in hci_sock_bind().
>
> >
> > That way it is simpler to protect the likes of
> > copy_from_user/copy_to_user, etc, even if we have to some checks
> > duplicated on each function we can have a helper function to checks
> > the flags, etc.
>
> My patch calls copy_from_user()/copy_to_user() without lock_sock()
> which works nicely with "[PATCH v3] Bluetooth: call lock_sock() outside
> of spinlock section". I'd like to backport "[PATCH v2] Bluetooth:
> reorganize ioctls from hci_sock_bound_ioctl()" together.

Yep, Im not asking you to change any of that.


-- 
Luiz Augusto von Dentz