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  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Sun, 6 Nov 2016 15:52:42 +0100
From:   Hans de Goede <hdegoede@...hat.com>
To:     Jacek Anaszewski <j.anaszewski@...sung.com>,
        linux-leds@...r.kernel.org
Cc:     linux-kernel@...r.kernel.org,
        Sakari Ailus <sakari.ailus@...ux.intel.com>,
        Pavel Machek <pavel@....cz>, Andrew Lunn <andrew@...n.ch>
Subject: Re: [PATCH] leds: Add mutex protection in brightness_show()

Hi,

On 04-11-16 17:46, Hans de Goede wrote:
> Hi,
>
> On 04-11-16 17:06, Jacek Anaszewski wrote:
>> Hi Hans,
>>
>> On 11/04/2016 12:53 PM, Hans de Goede wrote:
>>> Hi,
>>>
>>> On 04-11-16 08:52, Jacek Anaszewski wrote:
>>>> Initially the claim about no need for lock in brightness_show()
>>>> was valid as the function was just returning unchanged
>>>> LED brightness. After the addition of led_update_brightness() this
>>>> is no longer true, as the function can change the brightness if
>>>> a LED class driver implements brightness_get op. It can lead to
>>>> races between led_update_brightness() and led_set_brightness(),
>>>> resulting in overwriting new brightness with the old one before
>>>> the former is written to the device.
>>>>
>>>> Signed-off-by: Jacek Anaszewski <j.anaszewski@...sung.com>
>>>> Cc: Hans de Goede <hdegoede@...hat.com>
>>>> Cc: Sakari Ailus <sakari.ailus@...ux.intel.com>
>>>> Cc: Pavel Machek <pavel@....cz>
>>>> Cc: Andrew Lunn <andrew@...n.ch>
>>>> ---
>>>>  drivers/leds/led-class.c | 3 ++-
>>>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/leds/led-class.c b/drivers/leds/led-class.c
>>>> index 731e4eb..0c2307b 100644
>>>> --- a/drivers/leds/led-class.c
>>>> +++ b/drivers/leds/led-class.c
>>>> @@ -30,8 +30,9 @@ static ssize_t brightness_show(struct device *dev,
>>>>  {
>>>>      struct led_classdev *led_cdev = dev_get_drvdata(dev);
>>>>
>>>> -    /* no lock needed for this */
>>>> +    mutex_lock(&led_cdev->led_access);
>>>>      led_update_brightness(led_cdev);
>>>> +    mutex_unlock(&led_cdev->led_access);
>>>>
>>>>      return sprintf(buf, "%u\n", led_cdev->brightness);
>>>>  }
>>>>
>>>
>>> I'm afraid that this fix is not enough, the led_access lock is only
>>> held when the brightness is being updated through sysfs, not for
>>> trigger / sw-blinking updates (which cannot take a mutex as they
>>> may be called from non blocking contexts).
>>>
>>> We may need to consider to add a spinlock to the led_classdev and
>>> always lock that when calling into the driver, except for when
>>> the driver has a brightness_set_blocking callback. Which will need
>>> special handling.
>>
>> led_update_brightness() currently has two users besides LED subsystem
>> (at least grep reports those places):
>>
>> 1. v4l2-flash-led-class wrapper, for which led_access mutex was
>>    introduced. Its purpose was to disable LED sysfs interface while
>>    v4l2-flash wrapper takes over control of LED class device
>>    (not saying that the mutex wasn't missing even without this
>>     use case). Now I've realized that the call to
>>     led_sysfs_is_disabled() is missing in this patch.
>> 2. /drivers/platform/x86/thinkpad_acpi.c - it calls
>>    led_update_brightness() on suspend
>>
>> I think that the best we can now do is to add
>> lockdep_assert_held(&led_cdev->led_access) in led_update_brightness()
>> and a description saying that the caller has to acquire led_access
>> lock before calling it. Similarly as in case of
>> led_sysfs_disable()/led_sysfs_disable().
>
> The problem is not only callers of led_update_brightness() not holding
> led_cdev->led_access, the problem is also callers of led_set_brightness
> not holding led_cdev->led_access and they cannot take this lock because
> they may be called from a non-blocking context.

Thinking more about this, using a spinlock is also not going to work
because led_cdev->brightness_set_blocking and led_cdev->brightness_get
can both block and thus cannot be called with a spinlock held.

I think that we need to just make this a problem of the led drivers
and in include/linux/leds.h document that the led-core does not do
locking and that the drivers themselves need to protect against
their brightness_set / brightness_get callbacks when necessary.

Regards,

Hans

Powered by blists - more mailing lists