Date:   Wed, 30 Jan 2019 11:07:38 +0100
From:   Johan Hovold <johan@kernel.org>
To:     Myungho Jung <mhjungk@gmail.com>
Cc:     Marcel Holtmann <marcel@holtmann.org>,
        Johan Hedberg <johan.hedberg@gmail.com>,
        linux-bluetooth@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] Bluetooth: hci_uart: Switch pty driver to slave side in
 tty_set_termios()
Message-ID: <20190130100738.GQ3691@localhost>
References: <20190128065259.GA2976@myunghoj-Precision-5530>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190128065259.GA2976@myunghoj-Precision-5530>
User-Agent: Mutt/1.11.2 (2019-01-07)
Sender: linux-bluetooth-owner@vger.kernel.org
Precedence: bulk

On Sun, Jan 27, 2019 at 10:53:02PM -0800, Myungho Jung wrote:
> tty_set_termios() should be called with slave side of pty driver. So, If
> tty driver is pty master, it needs to be switched to ->link.

I'm not sure that's the right solution. PTYs are virtual devices used
for IPC and neither end (master or slave) have support for modem
control or baud rates.

> Reported-by: syzbot+a950165cbb86bdd023a4@syzkaller.appspotmail.com
> Signed-off-by: Myungho Jung <mhjungk@gmail.com>
> ---
>  drivers/bluetooth/hci_ldisc.c | 20 +++++++++++++++-----
>  1 file changed, 15 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/bluetooth/hci_ldisc.c b/drivers/bluetooth/hci_ldisc.c
> index fbf7b4df23ab..90c5ea8c399b 100644
> --- a/drivers/bluetooth/hci_ldisc.c
> +++ b/drivers/bluetooth/hci_ldisc.c
> @@ -299,10 +299,18 @@ static int hci_uart_send_frame(struct hci_dev *hdev, struct sk_buff *skb)
>  	return 0;
>  }
>  
> +/* If driver is pty master, return slave side */
> +static struct tty_struct *hci_uart_get_real_tty(struct tty_struct *tty)
> +{
> +	return  (tty->driver->type == TTY_DRIVER_TYPE_PTY &&
> +		 tty->driver->subtype == PTY_TYPE_MASTER) ? tty->link : tty;
> +}
> +
>  /* Flow control or un-flow control the device */
>  void hci_uart_set_flow_control(struct hci_uart *hu, bool enable)
>  {
>  	struct tty_struct *tty = hu->tty;
> +	struct tty_struct *real_tty;
>  	struct ktermios ktermios;
>  	int status;
>  	unsigned int set = 0;
> @@ -314,11 +322,12 @@ void hci_uart_set_flow_control(struct hci_uart *hu, bool enable)
>  		return;
>  	}
>  
> +	real_tty = hci_uart_get_real_tty(tty);
>  	if (enable) {
>  		/* Disable hardware flow control */
> -		ktermios = tty->termios;
> +		ktermios = real_tty->termios;
>  		ktermios.c_cflag &= ~CRTSCTS;
> -		status = tty_set_termios(tty, &ktermios);
> +		status = tty_set_termios(real_tty, &ktermios);
>  		BT_DBG("Disabling hardware flow control: %s",
>  		       status ? "failed" : "success");

So instead of these pointless calls to set the slave termios and
modem-control state, you might as well bail out early above (and
similarly in set_baudrate()).

Using n_hci for a master pty really makes no sense at all, so we could
even bail out at ldisc open, but perhaps that can be discussed and
addressed later.

Johan