[PATCH] Bluetooth: rfcomm: Release BTM while sleeping to avoid deadlock.

[PATCH] Bluetooth: rfcomm: Release BTM while sleeping to avoid deadlock.

[Владислав]

Bug description:
Kernel freezes for ~2 seconds after trying to access /dev/rfcommX device.
Steps to reproduce:
1. Edit /etc/bluetooth/rfcomm.conf - uncomment default config, set "bind" to "yes".
2. Issue "sudo rfcomm bind all" to create /dev/rfcomm0 device node.
3. Issue "cat /dev/rfcomm0" to trigger a bug.
Everything freezes (at least GUI): cursor does not move, no response from
keypresses, no gui redraw for 2~3 seconds.
Then cat displays expected error message "cat: /dev/rfcomm0: Host is down"
and system starts to respond again.
What expected (and occurs with 2.6.35) : no freeze between cat and error message.
Things goes worse when device, mentioned in /etc/bluetooth/rfcomm.conf
is in range and not paired - system displays pairing request and locks
up for a longer time, maybe completely - I did not wait longer than ~30
seconds and turned off bluetooth device.

Fix:
Release BTM while sleeping to avoid deadlock.

diff --git a/net/bluetooth/rfcomm/tty.c b/net/bluetooth/rfcomm/tty.c
index 2575c2d..d7b9af4 100644
--- a/net/bluetooth/rfcomm/tty.c
+++ b/net/bluetooth/rfcomm/tty.c
@@ -727,7 +727,9 @@ static int rfcomm_tty_open(struct tty_struct *tty, struct file *filp)
 			break;
 		}
 
+		tty_unlock();
 		schedule();
+		tty_lock();
 	}
 	set_current_state(TASK_RUNNING);
 	remove_wait_queue(&dev->wait, &wait);

Re: [PATCH] Bluetooth: rfcomm: Release BTM while sleeping to avoid deadlock.

[Gustavo F. Padovan]

Hi 

* Владислав <vladisslav@inbox.ru> [2011-02-13 04:01:45 +0300]:

> Bug description:
> Kernel freezes for ~2 seconds after trying to access /dev/rfcommX device.
> Steps to reproduce:
> 1. Edit /etc/bluetooth/rfcomm.conf - uncomment default config, set "bind" to "yes".
> 2. Issue "sudo rfcomm bind all" to create /dev/rfcomm0 device node.
> 3. Issue "cat /dev/rfcomm0" to trigger a bug.
> Everything freezes (at least GUI): cursor does not move, no response from
> keypresses, no gui redraw for 2~3 seconds.
> Then cat displays expected error message "cat: /dev/rfcomm0: Host is down"
> and system starts to respond again.
> What expected (and occurs with 2.6.35) : no freeze between cat and error message.
> Things goes worse when device, mentioned in /etc/bluetooth/rfcomm.conf
> is in range and not paired - system displays pairing request and locks
> up for a longer time, maybe completely - I did not wait longer than ~30
> seconds and turned off bluetooth device.
> 
> Fix:
> Release BTM while sleeping to avoid deadlock.
> 
> diff --git a/net/bluetooth/rfcomm/tty.c b/net/bluetooth/rfcomm/tty.c
> index 2575c2d..d7b9af4 100644
> --- a/net/bluetooth/rfcomm/tty.c
> +++ b/net/bluetooth/rfcomm/tty.c
> @@ -727,7 +727,9 @@ static int rfcomm_tty_open(struct tty_struct *tty, struct file *filp)
>  			break;
>  		}
>  
> +		tty_unlock();
>  		schedule();
> +		tty_lock();
>  	}
>  	set_current_state(TASK_RUNNING);
>  	remove_wait_queue(&dev->wait, &wait);

I've faced this issue some days ago and your patch looks good. But I need
proper git patch with your Signed-off-by line.

-- 
Gustavo F. Padovan
http://profusion.mobi

Re: [PATCH] Bluetooth: rfcomm: Release BTM while sleeping to avoid deadlock.

[Gustavo F. Padovan]

Hi Vladislav,

* Vladislav P <vladisslav@inbox.ru> [2011-02-14 18:50:07 +0300]:

> Mon, 14 Feb 2011 12:49:31 -0200 письмо от "Gustavo F. Padovan" <padovan@profusion.mobi>:
> 
> > Hi 
> > 
> > * Владислав <vladisslav@inbox.ru> [2011-02-13 04:01:45 +0300]:
> > 
> > > Bug description:
> > > Kernel freezes for ~2 seconds after trying to access /dev/rfcommX device.
> > > Steps to reproduce:
> > > 1. Edit /etc/bluetooth/rfcomm.conf - uncomment default config, set "bind" to
> > "yes".
> > > 2. Issue "sudo rfcomm bind all" to create /dev/rfcomm0 device node.
> > > 3. Issue "cat /dev/rfcomm0" to trigger a bug.
> > > Everything freezes (at least GUI): cursor does not move, no response from
> > > keypresses, no gui redraw for 2~3 seconds.
> > > Then cat displays expected error message "cat: /dev/rfcomm0: Host is down"
> > > and system starts to respond again.
> > > What expected (and occurs with 2.6.35) : no freeze between cat and error
> > message.
> > > Things goes worse when device, mentioned in /etc/bluetooth/rfcomm.conf
> > > is in range and not paired - system displays pairing request and locks
> > > up for a longer time, maybe completely - I did not wait longer than ~30
> > > seconds and turned off bluetooth device.
> > 
> > I've faced this issue some days ago and your patch looks good. But I need
> > proper git patch with your Signed-off-by line.
> > 
> > -- 
> > Gustavo F. Padovan
> > http://profusion.mobi
> Made proper patch:
> 
> From d1459f3042b4b8cdea1f1a184eb61f7be4002203 Mon Sep 17 00:00:00 2001
> From: Vladislav P <vladisslav@inbox.ru>
> Date: Mon, 14 Feb 2011 18:42:27 +0300
> Subject: [PATCH] Release BTM while sleeping to avoid deadlock.
> 
> Signed-off-by: Vladislav P <vladisslav@inbox.ru>
> ---
>  net/bluetooth/rfcomm/tty.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)

Applied, thanks.

-- 
Gustavo F. Padovan
http://profusion.mobi