MIME-Version: 1.0
References: <20230119013405.3870506-1-iam@sung-woo.kim> <CANn89iK6DZodENC8pR-toW_n5-VFyQR8X1XOuG9Lx1-kr1tmqQ@mail.gmail.com>
In-Reply-To: <CANn89iK6DZodENC8pR-toW_n5-VFyQR8X1XOuG9Lx1-kr1tmqQ@mail.gmail.com>
From:   Luiz Augusto von Dentz <luiz.dentz@gmail.com>
Date:   Fri, 20 Jan 2023 14:09:28 -0800
Message-ID: <CABBYNZLifwiJdeKmH4Abpe_uU_1BaCCPhuaUH=DtGLqGNFHHpQ@mail.gmail.com>
Subject: Re: [PATCH] L2CAP: Fix null-ptr-deref in l2cap_sock_set_shutdown_cb
To:     Eric Dumazet <edumazet@google.com>
Cc:     Sungwoo Kim <iam@sung-woo.kim>, daveti@purdue.edu, wuruoyu@me.com,
        benquike@gmail.com, Marcel Holtmann <marcel@holtmann.org>,
        Johan Hedberg <johan.hedberg@gmail.com>,
        "David S. Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        "open list:BLUETOOTH SUBSYSTEM" <linux-bluetooth@vger.kernel.org>,
        "open list:NETWORKING [GENERAL]" <netdev@vger.kernel.org>,
        open list <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Kim, Eric,

On Wed, Jan 18, 2023 at 8:16 PM Eric Dumazet <edumazet@google.com> wrote:
>
> On Thu, Jan 19, 2023 at 2:35 AM Sungwoo Kim <iam@sung-woo.kim> wrote:
> >
> > The L2CAP socket shutdown invokes l2cap_sock_destruct without a lock
> > on conn->chan_lock, assigning NULL to chan->data *just before*
> > the l2cap_disconnect_req thread that accesses to chan->data.
>
> This is racy then ?
>
> > This patch prevent it by adding a null check for a workaround, instead
> > of fixing a lock.
>
> This would at least require some barriers I think.
>
> What about other _cb helpers also reading/using chan->data ?

Perhaps it would be a good idea to include the stack backtrace so we
can better understand it, at some point we might need to refactor or
locks to avoid circular dependencies.

> >
> > This bug is found by FuzzBT, a modified Syzkaller by Sungwoo Kim(me).
> > Ruoyu Wu(wuruoyu@me.com) and Hui Peng(benquike@gmail.com) has helped
> > the FuzzBT project.
> >
> > Signed-off-by: Sungwoo Kim <iam@sung-woo.kim>
>
> I would also add
>
> Fixes: 1bff51ea59a9 ("Bluetooth: fix use-after-free error in
> lock_sock_nested()")

+1

> > ---
> >  net/bluetooth/l2cap_sock.c | 8 +++++---
> >  1 file changed, 5 insertions(+), 3 deletions(-)
> >
> > diff --git a/net/bluetooth/l2cap_sock.c b/net/bluetooth/l2cap_sock.c
> > index ca8f07f35..350c7afdf 100644
> > --- a/net/bluetooth/l2cap_sock.c
> > +++ b/net/bluetooth/l2cap_sock.c
> > @@ -1681,9 +1681,11 @@ static void l2cap_sock_set_shutdown_cb(struct l2cap_chan *chan)
> >  {
> >         struct sock *sk = chan->data;
> >
>
> Other similar fixes simply do:
>
>      if (!sk)
>           return;
>
> I would chose to use the same coding style in net/bluetooth/l2cap_sock.c

Yep, at least l2cap_sock_close_cb and l2cap_sock_shutdown do that already.

>
> > -       lock_sock(sk);
> > -       sk->sk_shutdown = SHUTDOWN_MASK;
> > -       release_sock(sk);
> > +       if (!sk) {
> > +               lock_sock(sk);
> > +               sk->sk_shutdown = SHUTDOWN_MASK;
> > +               release_sock(sk);
> > +       }
> >  }
> >
> >  static long l2cap_sock_get_sndtimeo_cb(struct l2cap_chan *chan)
> > --
> > 2.25.1
> >


-- 
Luiz Augusto von Dentz