Re: [PATCH v9] leds: USB: HID: Add support for MSI GT683R led panels

[Greg KH <greg@kroah.com>]

On Mon, Jun 23, 2014 at 04:42:55PM +0200, Johan Hovold wrote:
> On Mon, Jun 23, 2014 at 04:35:13PM +0200, Oliver Neukum wrote:
> > On Wed, 2014-06-18 at 19:05 +0300, Janne Kanniainen wrote:
> > > This driver adds support for USB controlled led panels that exists in
> > > MSI GT683R laptop
> > > 
> > 
> > > +static int gt683r_led_probe(struct hid_device *hdev,
> > > +			const struct hid_device_id *id)
> > > +{
> > > +	int i;
> > > +	int ret;
> > > +	int name_sz;
> > > +	char *name;
> > > +	struct gt683r_led *led;
> > > +
> > > +	led = devm_kzalloc(&hdev->dev, sizeof(*led), GFP_KERNEL);
> > > +	if (!led)
> > > +		return -ENOMEM;
> > > +
> > > +	led->mode = GT683R_LED_NORMAL;
> > > +	led->hdev = hdev;
> > > +	hid_set_drvdata(hdev, led);
> > > +
> > > +	ret = hid_parse(hdev);
> > > +	if (ret) {
> > > +		hid_err(hdev, "hid parsing failed\n");
> > > +		return ret;
> > > +	}
> > > +
> > > +	ret = hid_hw_start(hdev, HID_CONNECT_HIDRAW);
> > > +	if (ret) {
> > > +		hid_err(hdev, "hw start failed\n");
> > > +		return ret;
> > > +	}
> > > +
> > > +	for (i = 0; i < GT683R_LED_COUNT; i++) {
> > > +		name_sz = strlen(dev_name(&hdev->dev)) +
> > > +				strlen(gt683r_panel_names[i]) + 3;
> > > +
> > > +		name = devm_kzalloc(&hdev->dev, name_sz, GFP_KERNEL);
> > > +		if (!name) {
> > > +			ret = -ENOMEM;
> > > +			goto fail;
> > > +		}
> > > +
> > > +		snprintf(name, name_sz, "%s::%s",
> > > +				dev_name(&hdev->dev), gt683r_panel_names[i]);
> > > +		led->led_devs[i].name = name;
> > > +		led->led_devs[i].max_brightness = 1;
> > > +		led->led_devs[i].brightness_set = gt683r_brightness_set;
> > > +		ret = led_classdev_register(&hdev->dev, &led->led_devs[i]);
> > > +		if (ret) {
> > > +			hid_err(hdev, "could not register led device\n");
> > > +			goto fail;
> > > +		}
> > > +	}
> > > +
> > > +	ret = device_create_file(&led->hdev->dev, &dev_attr_leds_mode);
> > > +	if (ret) {
> > > +		hid_err(hdev, "could not make mode attribute file\n");
> > > +		goto fail;
> > > +	}
> > > +
> > 
> > This is the window.
> > 
> > > +	mutex_init(&led->lock);
> > > +	INIT_WORK(&led->work, gt683r_led_work);
> > > +
> > 
> > And here we have a problem. This is a race condition.
> > At this time you've already created the sysfs files. So
> > their methods can be called. They will lock a mutex and/or
> > schedule work that hasn't been initialized.
> > The initialization must be done before anything in sysfs
> > is created. 
> 
> Just move the initialisation of the lock and work to the other private
> data initialisations directly after it's allocated.
> 
> Can you send a follow up patch, Janne?

No driver should be creating sysfs files directly, use the proper
attribute group interface instead.  That way the files are created
_before_ the device is announced to userspace, instead of after, which
is a race condition there as well.

thanks,

greg k-h