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: <20201030160833.BlueZ.v1.1.Ia45f3edc48142d9db0dc4b315c84ab60a149697f@changeid>
In-Reply-To: <20201030160833.BlueZ.v1.1.Ia45f3edc48142d9db0dc4b315c84ab60a149697f@changeid>
From:   Luiz Augusto von Dentz <luiz.dentz@gmail.com>
Date:   Fri, 30 Oct 2020 16:48:05 -0700
Message-ID: <CABBYNZKck6MFew6r-pEJgh2CLWUNmrKf4KO5z-nzBVJUBVcSmQ@mail.gmail.com>
Subject: Re: [BlueZ PATCH v1] adapter: Fix a crash caused by lingering
 discovery client pointer
To:     Miao-chen Chou <mcchou@chromium.org>
Cc:     Bluetooth Kernel Mailing List <linux-bluetooth@vger.kernel.org>,
        Marcel Holtmann <marcel@holtmann.org>,
        Alain Michaud <alainm@chromium.org>,
        Sonny Sasaka <sonnysasaka@chromium.org>,
        ChromeOS Bluetooth Upstreaming 
        <chromeos-bluetooth-upstreaming@chromium.org>,
        Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Miao-chen,

On Fri, Oct 30, 2020 at 4:13 PM Miao-chen Chou <mcchou@chromium.org> wrote:
>
> This cleans up the lingering pointer, adapter->client, during powering
> off the adapter. The crash occurs when a D-Bus client set Powered
> property to false and immediately calls StopDiscovery() when there is
> ongoing discovery. As a part of powering off the adapter,
> adapter->discovery_list gets cleared, and given that adapter->client
> refers to one of the clients in adapter->discovery_list, adapter->client
> should be cleared along with it.
>
> Reviewed-by: Alain Michaud <alainm@chromium.org>
> Reviewed-by: Sonny Sasaka <sonnysasaka@chromium.org>
> ---
>
>  src/adapter.c | 17 ++++++++++++++++-
>  1 file changed, 16 insertions(+), 1 deletion(-)
>
> diff --git a/src/adapter.c b/src/adapter.c
> index c0053000a..74bfb0448 100644
> --- a/src/adapter.c
> +++ b/src/adapter.c
> @@ -1507,8 +1507,10 @@ static void discovery_free(void *user_data)
>                 client->discovery_filter = NULL;
>         }
>
> -       if (client->msg)
> +       if (client->msg) {
>                 dbus_message_unref(client->msg);
> +               client->msg = NULL;
> +       }

This shouldn't make any difference as the whole list is freed and so
is the client.

>
>         g_free(client->owner);
>         g_free(client);
> @@ -5301,6 +5303,19 @@ static void free_service_auth(gpointer data, gpointer user_data)
>
>  static void remove_discovery_list(struct btd_adapter *adapter)
>  {
> +       DBusMessage *msg;
> +
> +       if (adapter->client) {
> +               msg = adapter->client->msg;
> +               if (msg) {
> +                       g_dbus_send_message(dbus_conn, btd_error_busy(msg));
> +                       dbus_message_unref(msg);
> +                       adapter->client->msg = NULL;
> +               }
> +
> +               adapter->client = NULL;

Shouldn't you call discovery_free as well here? Or perhaps we could
move the lines above inside discovery_free so it detects if the
adapter->client is pointing to a client that is being freed.

> +       }
> +
>         g_slist_free_full(adapter->set_filter_list, discovery_free);
>         adapter->set_filter_list = NULL;
>
> --
> 2.26.2
>


-- 
Luiz Augusto von Dentz