leds: trigger: ledtrig-backlight: Fix invalid memory access in fb_event notification callback
PATCH 1/1 leds: trigger: ledtrig-backlight: Fix invalid memory access in fb_event notification callback
fb_notifier_callback is called on any event fired by fb_notifier_call_chain. Events may, or may not contain some data (fb_event.data).
In case of FB_EVENT_BLANK fb_event.data contains a pointer to an integer holding the blank state. The Problem is, that in ledtrig-backlight.c - fb_notifier_callback the pointer to blank state is dereferenced BEFORE the event-type is checked.
Obviously this leads to problems with other events than FB_EVENT_BLANK, where fb_event.data is undefined or NULL.
It seems, that this problem existed ever since the driver was added.
Like in drivers/video/backlight/backlight.c line 43 I would suggest to return immediately on events other than FB_EVENT_BLANK.
---
Background information:
I'm currently working on a IMX53(ARM) based hardware and Linux 3.11-rc5 and detected a problem in drivers/leds/trigger/ledtrig-backlight.c
ledtrig is used on our hardware (besides pwm-backlight) to switch off a gpio-line connected to display-supply enable-line, when the display is blanked.
On first boot after configuring ledtrig I detected an invalid memory access in ledtrig-backlight.c - fb_notifier_callback. In my special case the segfault was caused by an FB_EVENT_FB_REGISTERED, fired with undefined fb_event.data in drivers/video/fbmem.c line 1653.
drivers/leds/trigger/ledtrig-backlight.c +17 -15
@@ -36,26 +36,28 @@ static int fb_notifier_callback(struct notifier_block *p, struct bl_trig_notifier, notifier);struct led_classdev *led = n->led;struct fb_event *fb_event = data; -int *blank = fb_event->data; -int new_status = *blank ?BLANK :UNBLANK ; +int *blank; +int new_status; + +/* If we aren't interested in this event, skip it immediately ... */ +if (event !=FB_EVENT_BLANK ) +return 0 ; -switch (event) { -case FB_EVENT_BLANK : -if (new_status == n->old_status) -break ; + blank = fb_event->data; + new_status = *blank ?BLANK :UNBLANK ; -if ((n->old_status ==UNBLANK ) ^ n->invert) { - n->brightness = led->brightness; -__led_set_brightness (led,LED_OFF ); - }else { -__led_set_brightness (led, n->brightness); - } +if (new_status == n->old_status) +return 0 ; - n->old_status = new_status; - -break ; +if ((n->old_status ==UNBLANK ) ^ n->invert) { + n->brightness = led->brightness; +__led_set_brightness (led,LED_OFF ); + }else { +__led_set_brightness (led, n->brightness); } + n->old_status = new_status; +return 0 ; }
Quoted 8 lines · Manfred Schlaegl
fb_notifier_callbackis called on any event fired byfb_notifier_call_chain. Events may, or may not contain some data (fb_event.data).
In case ofFB_EVENT_BLANKfb_event.data contains a pointer to an integer holding the blank state. The Problem is, that inledtrig-backlight.c-fb_notifier_callbackthe pointer to blank state is dereferenced BEFORE the event-type is checked.
Obviously this leads to problems with other events thanFB_EVENT_BLANK, where fb_event.data is undefined or NULL.
It seems, that this problem existed ever since the driver was added.Like in
drivers/video/backlight/backlight.cline 43 I would suggest to return immediately on events other thanFB_EVENT_BLANK.
Indeed, this is an issue. Thanks for this nice catch and fixing.
I merged this patch into my tree and it will hit 3.11 release.
Thanks,
-Bryan
Quoted 61 lines
Signed-off-by Manfred Schlaegl <[email protected]>---
Background information:
I'm currently working on a IMX53(ARM) based hardware and Linux 3.11-rc5 and detected a problem indrivers/leds/trigger/ledtrig-backlight.c
ledtrig is used on our hardware (besides pwm-backlight) to switch off a gpio-line connected to display-supply enable-line, when the display is blanked.
On first boot after configuring ledtrig I detected an invalid memory access inledtrig-backlight.c-fb_notifier_callback. In my special case the segfault was caused by anFB_EVENT_FB_REGISTERED, fired with undefined fb_event.data indrivers/video/fbmem.cline 1653.drivers/leds/trigger/ledtrig-backlight.c +17
-15@@ -36,26 +36,28 @@ static int fb_notifier_callback(struct notifier_block *p, struct bl_trig_notifier, notifier);struct led_classdev *led = n->led;struct fb_event *fb_event = data; -int *blank = fb_event->data; -int new_status = *blank ?BLANK :UNBLANK ; +int *blank; +int new_status; + +/* If we aren't interested in this event, skip it immediately ... */ +if (event !=FB_EVENT_BLANK ) +return 0 ; -switch (event) { -case FB_EVENT_BLANK : -if (new_status == n->old_status) -break ; + blank = fb_event->data; + new_status = *blank ?BLANK :UNBLANK ; -if ((n->old_status ==UNBLANK ) ^ n->invert) { - n->brightness = led->brightness; -__led_set_brightness (led,LED_OFF ); - }else { -__led_set_brightness (led, n->brightness); - } +if (new_status == n->old_status) +return 0 ; - n->old_status = new_status; - -break ; +if ((n->old_status ==UNBLANK ) ^ n->invert) { + n->brightness = led->brightness; +__led_set_brightness (led,LED_OFF ); + }else { +__led_set_brightness (led, n->brightness); } + n->old_status = new_status; +return 0 ; }
Quoted 11 lines · Bryan Wu
Quoted 8 lines · Manfred Schlaegl
fb_notifier_callbackis called on any event fired byfb_notifier_call_chain. Events may, or may not contain some data (fb_event.data).
In case ofFB_EVENT_BLANKfb_event.data contains a pointer to an integer holding the blank state. The Problem is, that inledtrig-backlight.c-fb_notifier_callbackthe pointer to blank state is dereferenced BEFORE the event-type is checked.
Obviously this leads to problems with other events thanFB_EVENT_BLANK, where fb_event.data is undefined or NULL.
It seems, that this problem existed ever since the driver was added.Like in
drivers/video/backlight/backlight.cline 43 I would suggest to return immediately on events other thanFB_EVENT_BLANK.
Quoted 6 lines
Indeed, this is an issue. Thanks for this nice catch and fixing.
I merged this patch into my tree and it will hit 3.11 release.
Thanks,
-Bryan
That would be great.
Thanks too.
-manfred
Quoted 67 lines
Quoted 52 lines
Signed-off-by Manfred Schlaegl <[email protected]>---
Background information:
I'm currently working on a IMX53(ARM) based hardware and Linux 3.11-rc5 and detected a problem indrivers/leds/trigger/ledtrig-backlight.c
ledtrig is used on our hardware (besides pwm-backlight) to switch off a gpio-line connected to display-supply enable-line, when the display is blanked.
On first boot after configuring ledtrig I detected an invalid memory access inledtrig-backlight.c-fb_notifier_callback. In my special case the segfault was caused by anFB_EVENT_FB_REGISTERED, fired with undefined fb_event.data indrivers/video/fbmem.cline 1653.drivers/leds/trigger/ledtrig-backlight.c +12
-15@@ -36,26 +36,28 @@ static int fb_notifier_callback(struct notifier_block *p, struct bl_trig_notifier, notifier);struct led_classdev *led = n->led;struct fb_event *fb_event = data; -int *blank = fb_event->data; -int new_status = *blank ?BLANK :UNBLANK ; +int *blank; +int new_status; + +/* If we aren't interested in this event, skip it immediately ... */ +if (event !=FB_EVENT_BLANK ) +return 0 ; -switch (event) { -case FB_EVENT_BLANK : -if (new_status == n->old_status) -break ; + blank = fb_event->data; + new_status = *blank ?BLANK :UNBLANK ; -if ((n->old_status ==UNBLANK ) ^ n->invert) { - n->brightness = led->brightness; -__led_set_brightness (led,LED_OFF ); - }else { -__led_set_brightness (led, n->brightness); - } +if (new_status == n->old_status) +return 0 ; - n->old_status = new_status; - -break ; +if ((n->old_status ==UNBLANK ) ^ n->invert) { + n->brightness = led->brightness;drivers/leds/trigger/ledtrig-backlight.c static int fb_notifier_callback(struct notifier_block *p, · line 63
Quoted 10 lines
drivers/leds/trigger/ledtrig-backlight.c +5
-0+__led_set_brightness (led,LED_OFF ); + }else { +__led_set_brightness (led, n->brightness); } + n->old_status = new_status; +return 0 ; }--
To unsubscribe from this list: send the line "unsubscribe linux-leds" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html