linux-kernel - Re: [PATCH v2] Bluetooth: hci_core: fix init for HCI_USER

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-Id: <0B1D4151-BB95-47A6-B062-686DF7362A6B@holtmann.org>
Date:   Thu, 17 Oct 2019 07:11:57 +0200
From:   Marcel Holtmann <marcel@...tmann.org>
To:     Mattijs Korpershoek <mkorpershoek@...libre.com>
Cc:     Bluez mailing list <linux-bluetooth@...r.kernel.org>,
        Johan Hedberg <johan.hedberg@...il.com>,
        "David S. Miller" <davem@...emloft.net>, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] Bluetooth: hci_core: fix init for HCI_USER_CHANNEL

Hi Mattijs,

> During the setup() stage, HCI device drivers expect the chip to
> acknowledge its setup() completion via vendor specific frames.
> 
> If userspace opens() such HCI device in HCI_USER_CHANNEL [1] mode,
> the vendor specific frames are never tranmitted to the driver, as
> they are filtered in hci_rx_work().
> 
> Allow HCI devices which operate in HCI_USER_CHANNEL mode to receive
> frames if the HCI device is is HCI_INIT state.
> 
> [1] https://www.spinics.net/lists/linux-bluetooth/msg37345.html
> 
> Fixes: 23500189d7e0 ("Bluetooth: Introduce new HCI socket channel for user operation")
> Signed-off-by: Mattijs Korpershoek <mkorpershoek@...libre.com>
> ---
> Changelog:
> v2:
> * change test logic to transfer packets when in INIT phase
>  for user channel mode as recommended by Marcel
> * renamed patch from
>  "Bluetooth: hci_core: fix init with HCI_QUIRK_NON_PERSISTENT_SETUP"
> 
> v1:
> * https://lkml.org/lkml/2019/10/3/2250
> 
> Some more background on the change follows:
> 
> The Android bluetooth stack (Bluedroid) also has a HAL implementation
> which follows Linux's standard rfkill interface [1].
> 
> This implementation relies on the HCI_CHANNEL_USER feature to get
> exclusive access to the underlying bluetooth device.
> 
> When testing this along with the btkmtksdio driver, the
> chip appeared unresponsive when calling the following from userspace:
> 
>    struct sockaddr_hci addr;
>    int fd;
> 
>    fd = socket(AF_BLUETOOTH, SOCK_RAW, BTPROTO_HCI);
> 
>    memset(&addr, 0, sizeof(addr));
>    addr.hci_family = AF_BLUETOOTH;
>    addr.hci_dev = 0;
>    addr.hci_channel = HCI_CHANNEL_USER;
> 
>    bind(fd, (struct sockaddr *) &addr, sizeof(addr)); # device hangs
> 
> In the case of bluetooth drivers exposing QUIRK_NON_PERSISTENT_SETUP
> such as btmtksdio, setup() is called each multiple times.
> In particular, when userspace calls bind(), the setup() is called again
> and vendor specific commands might be send to re-initialize the chip.
> 
> Those commands are filtered out by hci_core in HCI_CHANNEL_USER mode,
> preventing setup() from completing successfully.
> 
> This has been tested on a 4.19 kernel based on Android Common Kernel.
> It has also been compile tested on bluetooth-next.
> 
> [1] https://android.googlesource.com/platform/system/bt/+/refs/heads/master/vendor_libs/linux/interface/
> 
> net/bluetooth/hci_core.c | 9 ++++++++-
> 1 file changed, 8 insertions(+), 1 deletion(-)

patch has been applied to bluetooth-next tree.

Regards

Marcel