Re: [BUG] ati_remote2.c: possible mutex_lock without mutex_unlock

[Jiri Kosina <jkosina@suse.cz>]

On Tue, 13 Oct 2009, Dmitry Torokhov wrote:

> Umm, I don't like assuming that EAGAIN can only mean that 
> mutex_lock_interruptible() failed, seq_file core may theoretically 
> return -EAGAIN too. In fact, looking through seq_file.c traverse() does 
> return -EAGAIN in certain cases...

Damn, you are right -- I explicitly checked for this, but have completely 
overlooked the "Eoveflow:" branch in traverse(), which returns EAGAIN. So 
my previous patch is of course incorrect.

> Input: fix locking issue in /proc/bus/input/ handlers
> 
> From: Dmitry Torokhov <dmitry.torokhov@gmail.com>
> 
> input_devices_seq_start() uses mutex_lock_interruptible() to acquire
> the input_mutex, but doesn't properly handle the situation when the
> call fails (for example due to interrupt). Instead of returning NULL
> (which indicates that there is no more data) we should return
> ERR_PTR()-encoded error.
> 
> We also need explicit flag indicating whether input_mutex was acquired
> since input_devices_seq_stop() is called whether input_devices_seq_start()
> was successful or not.
> 
> The same applies to input_handlers_seq_start().
> 
> Reported-by: iceberg <strakh@ispras.ru>
> Signed-off-by: Dmitry Torokhov <dtor@mail.ru>

Yup, looks OK to me.

-- 
Jiri Kosina
SUSE Labs, Novell Inc.