Re: [PATCH][next] drm/kmb: fix array out-of-bounds writes to kmb->plane_status[]

[Colin Ian King]

On 16/11/2020 16:53, Chrisanthus, Anitha wrote:
> Hi Sam and Colin,
> 
>> -----Original Message-----
>> From: Sam Ravnborg <sam@xxxxxxxxxxxx>
>> Sent: Friday, November 13, 2020 10:02 AM
>> To: Colin Ian King <colin.king@xxxxxxxxxxxxx>
>> Cc: Chrisanthus, Anitha <anitha.chrisanthus@xxxxxxxxx>; Dea, Edmund J
>> <edmund.j.dea@xxxxxxxxx>; David Airlie <airlied@xxxxxxxx>; Daniel Vetter
>> <daniel@xxxxxxxx>; dri-devel@xxxxxxxxxxxxxxxxxxxxx; kernel-
>> janitors@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
>> Subject: Re: [PATCH][next] drm/kmb: fix array out-of-bounds writes to kmb-
>>> plane_status[]
>>
>> Hi Colin.
>> On Fri, Nov 13, 2020 at 03:04:34PM +0000, Colin Ian King wrote:
>>> On 13/11/2020 14:55, Sam Ravnborg wrote:
>>>> Hi Colin.
>>>>
>>>> On Fri, Nov 13, 2020 at 12:01:21PM +0000, Colin King wrote:
>>>>> From: Colin Ian King <colin.king@xxxxxxxxxxxxx>
>>>>>
>>>>> Writes to elements in the kmb->plane_status array in function
>>>>> kmb_plane_atomic_disable are overrunning the array when plane_id is
>>>>> more than 1 because currently the array is KMB_MAX_PLANES elements
>>>>> in size and this is currently #defined as 1.  Fix this by defining
>>>>> KMB_MAX_PLANES to 4.
>>>>
>>>> I fail to follow you here.
>>>> In kmb_plane_init() only one plane is allocated - with id set to 0.
>>>> So for now only one plane is allocated thus kmb_plane_atomic_disable()
>>>> is only called for this plane.
>>>>
>>>> With your change we will start allocating four planes, something that is
>>>> not tested.
>>>>
>>>> Do I miss something?
>>>>
>>>> 	Sam
>>>>
>>>
>>> The static analysis from coverity on linux-next suggested that there was
>>> an array overflow as follows:
>>>
>>> 108 static void kmb_plane_atomic_disable(struct drm_plane *plane,
>>> 109                                     struct drm_plane_state *state)
>>> 110 {
>>>
>>>    1. Condition 0 /* !!(!__builtin_types_compatible_p() &&
>>> !__builtin_types_compatible_p()) */, taking false branch.
>>>
>>> 111        struct kmb_plane *kmb_plane = to_kmb_plane(plane);
>>>
>>>    2. assignment: Assigning: plane_id = kmb_plane->id.
>>>
>>> 112        int plane_id = kmb_plane->id;
>>> 113        struct kmb_drm_private *kmb;
>>> 114
>>> 115        kmb = to_kmb(plane->dev);
>>> 116
>>>
>>>    3. Switch case value LAYER_3.
>>>
>>> 117        switch (plane_id) {
>>> 118        case LAYER_0:
>>> 119                kmb->plane_status[plane_id].ctrl = LCD_CTRL_VL1_ENABLE;
>>> 120                break;
>>
>> With the current code this is the only case that hits.
>> So coverity is right that if we hit other cases that would result in a
>> bug. But kmb_plane->id will for now not have other values than 0.
>>
>> So it is a subtle false positive.
>> There is some "dead" code here - but this is in preparation for more
>> than one layer and we will keep the code for now, unless Anitha chimes
>> in and says otherwise.
> 
> Thanks Sam, I was out on Friday. Agree with Sam, let's keep the current code for now. Kmb->plane_id will not have non-zero values now.
> Only one plane is supported and tested currently, the extra code is in preparation for multiple planes.

Thanks for the clarification. Apologies for the noise.

> 
> Thanks,
> Anitha
>>
>> 	Sam
>>
>>> 121        case LAYER_1:
>>>
>>>    (#2 of 4): Out-of-bounds write (OVERRUN)
>>>
>>> 122                kmb->plane_status[plane_id].ctrl = LCD_CTRL_VL2_ENABLE;
>>> 123                break;
>>> 124        case LAYER_2:
>>>
>>>    (#3 of 4): Out-of-bounds write (OVERRUN)
>>>
>>> 125                kmb->plane_status[plane_id].ctrl = LCD_CTRL_GL1_ENABLE;
>>> 126                break;
>>>
>>>    4. equality_cond: Jumping to case LAYER_3.
>>>
>>> 127        case LAYER_3:
>>>
>>>    (#1 of 4): Out-of-bounds write (OVERRUN)
>>>    5. overrun-local: Overrunning array kmb->plane_status of 1 8-byte
>>> elements at element index 3 (byte offset 31) using index plane_id (which
>>> evaluates to 3).
>>>
>>> 128                kmb->plane_status[plane_id].ctrl = LCD_CTRL_GL2_ENABLE;
>>> 129                break;
>>> 130        }
>>> 131
>>>
>>>    (#4 of 4): Out-of-bounds write (OVERRUN)
>>>
>>> 132        kmb->plane_status[plane_id].disable = true;
>>> 133 }
>>> 134
>>>
>>> So it seems the assignments to  kmb->plane_status[plane_id] are
>>> overrunning the array since plane_status is allocated as 1 element and
>>> yet plane_id can be 0..3
>>>
>>> I could be misunderstanding this, or it may be a false positive.
>>>
>>> Colin