media: imx: csi: fix enum_mbus_code for unknown mbus format codes

Message ID 20180208144749.10558-1-p.zabel@pengutronix.de (mailing list archive)
State Changes Requested, archived
Headers

Commit Message

Philipp Zabel Feb. 8, 2018, 2:47 p.m. UTC
  If no imx_media_pixfmt is found for a given mbus format code,
we shouldn't crash. Return -EINVAL for any index.

Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
---
 drivers/staging/media/imx/imx-media-csi.c | 4 ++++
 1 file changed, 4 insertions(+)
  

Comments

Steve Longerbeam Feb. 8, 2018, 7:27 p.m. UTC | #1
On 02/08/2018 06:47 AM, Philipp Zabel wrote:
> If no imx_media_pixfmt is found for a given mbus format code,
> we shouldn't crash. Return -EINVAL for any index.

Hi Philipp,

If imx_media_find_mbus_format() returns NULL at that location, it means
the current format has an invalid pixel code. It's not possible for an 
ACTIVE
format to have an invalid code, so it must be a TRY format with an 
uninitialized
(zero) code. Which makes sense if enum_mbus_code(TRY) pad op is called 
before
set_fmt(TRY), at the sink pad, was *ever* called. Am I right?

It looks there is another location where this could possibly happen, in 
csi_try_fmt().
In that case it would happen if set_fmt(TRY) is called on a source pad, 
before
set_fmt(TRY) was ever called at the sink pad. That's a weird corner case 
because
it's not clear what a set_fmt(TRY) at the source pads should choose for 
pixel
code if there was never a set_fmt(TRY) at the sink pad.

But perhaps the following should be added to this patch as well. It 
makes the
assumption that the TRY code at the sink pad is the same as the default
active code set from csi_registered().

Or maybe I should ask the question, what should drivers do in set_fmt(TRY)
at their source pads, if there was no prior set_fmt(TRY) at their sink pads?

Steve


@@ -1281,6 +1281,11 @@ static void csi_try_fmt(struct csi_priv *priv,
         case CSI_SRC_PAD_IDMAC:
                 incc = imx_media_find_mbus_format(infmt->code,
                                                   CS_SEL_ANY, true);
+               if (!incc) {
+                       imx_media_enum_mbus_format(&code, 0, CS_SEL_YUV, 
false);
+                       incc = imx_media_find_mbus_format(code,
+ CS_SEL_YUV, false);
+               }

                 sdformat->format.width = compose->width;
                 sdformat->format.height = compose->height;



> Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
> ---
>   drivers/staging/media/imx/imx-media-csi.c | 4 ++++
>   1 file changed, 4 insertions(+)
>
> diff --git a/drivers/staging/media/imx/imx-media-csi.c b/drivers/staging/media/imx/imx-media-csi.c
> index eb7be5093a9d..89903f267d60 100644
> --- a/drivers/staging/media/imx/imx-media-csi.c
> +++ b/drivers/staging/media/imx/imx-media-csi.c
> @@ -1138,6 +1138,10 @@ static int csi_enum_mbus_code(struct v4l2_subdev *sd,
>   
>   	infmt = __csi_get_fmt(priv, cfg, CSI_SINK_PAD, code->which);
>   	incc = imx_media_find_mbus_format(infmt->code, CS_SEL_ANY, true);
> +	if (!incc) {
> +		ret = -EINVAL;
> +		goto out;
> +	}
>   
>   	switch (code->pad) {
>   	case CSI_SINK_PAD:
  
Hans Verkuil Feb. 8, 2018, 8:01 p.m. UTC | #2
On 02/08/2018 08:27 PM, Steve Longerbeam wrote:
> 
> 
> On 02/08/2018 06:47 AM, Philipp Zabel wrote:
>> If no imx_media_pixfmt is found for a given mbus format code,
>> we shouldn't crash. Return -EINVAL for any index.
> 
> Hi Philipp,
> 
> If imx_media_find_mbus_format() returns NULL at that location, it means
> the current format has an invalid pixel code. It's not possible for an 
> ACTIVE
> format to have an invalid code, so it must be a TRY format with an 
> uninitialized
> (zero) code. Which makes sense if enum_mbus_code(TRY) pad op is called 
> before
> set_fmt(TRY), at the sink pad, was *ever* called. Am I right?
> 
> It looks there is another location where this could possibly happen, in 
> csi_try_fmt().
> In that case it would happen if set_fmt(TRY) is called on a source pad, 
> before
> set_fmt(TRY) was ever called at the sink pad. That's a weird corner case 
> because
> it's not clear what a set_fmt(TRY) at the source pads should choose for 
> pixel
> code if there was never a set_fmt(TRY) at the sink pad.
> 
> But perhaps the following should be added to this patch as well. It 
> makes the
> assumption that the TRY code at the sink pad is the same as the default
> active code set from csi_registered().
> 
> Or maybe I should ask the question, what should drivers do in set_fmt(TRY)
> at their source pads, if there was no prior set_fmt(TRY) at their sink pads?

Drivers can set the initial TRY value by implementing the init_cfg pad op.
See e.g. drivers/media/platform/vimc/vimc-sensor.c.

I think you have two choices: either set it to some default format, or copy
the current ACTIVE format. It doesn't really matter, as long as there is
something valid.

Many drivers will likely fail this test.

Regards,

	Hans

> 
> Steve
> 
> 
> @@ -1281,6 +1281,11 @@ static void csi_try_fmt(struct csi_priv *priv,
>          case CSI_SRC_PAD_IDMAC:
>                  incc = imx_media_find_mbus_format(infmt->code,
>                                                    CS_SEL_ANY, true);
> +               if (!incc) {
> +                       imx_media_enum_mbus_format(&code, 0, CS_SEL_YUV, 
> false);
> +                       incc = imx_media_find_mbus_format(code,
> + CS_SEL_YUV, false);
> +               }
> 
>                  sdformat->format.width = compose->width;
>                  sdformat->format.height = compose->height;
> 
> 
> 
>> Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
>> ---
>>   drivers/staging/media/imx/imx-media-csi.c | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/drivers/staging/media/imx/imx-media-csi.c b/drivers/staging/media/imx/imx-media-csi.c
>> index eb7be5093a9d..89903f267d60 100644
>> --- a/drivers/staging/media/imx/imx-media-csi.c
>> +++ b/drivers/staging/media/imx/imx-media-csi.c
>> @@ -1138,6 +1138,10 @@ static int csi_enum_mbus_code(struct v4l2_subdev *sd,
>>   
>>   	infmt = __csi_get_fmt(priv, cfg, CSI_SINK_PAD, code->which);
>>   	incc = imx_media_find_mbus_format(infmt->code, CS_SEL_ANY, true);
>> +	if (!incc) {
>> +		ret = -EINVAL;
>> +		goto out;
>> +	}
>>   
>>   	switch (code->pad) {
>>   	case CSI_SINK_PAD:
>
  
Steve Longerbeam Feb. 10, 2018, 1:22 a.m. UTC | #3
Hi Hans,


On 02/08/2018 12:01 PM, Hans Verkuil wrote:
> On 02/08/2018 08:27 PM, Steve Longerbeam wrote:
>>
>> On 02/08/2018 06:47 AM, Philipp Zabel wrote:
>>> If no imx_media_pixfmt is found for a given mbus format code,
>>> we shouldn't crash. Return -EINVAL for any index.
>> Hi Philipp,
>>
>> If imx_media_find_mbus_format() returns NULL at that location, it means
>> the current format has an invalid pixel code. It's not possible for an
>> ACTIVE
>> format to have an invalid code, so it must be a TRY format with an
>> uninitialized
>> (zero) code. Which makes sense if enum_mbus_code(TRY) pad op is called
>> before
>> set_fmt(TRY), at the sink pad, was *ever* called. Am I right?
>>
>> It looks there is another location where this could possibly happen, in
>> csi_try_fmt().
>> In that case it would happen if set_fmt(TRY) is called on a source pad,
>> before
>> set_fmt(TRY) was ever called at the sink pad. That's a weird corner case
>> because
>> it's not clear what a set_fmt(TRY) at the source pads should choose for
>> pixel
>> code if there was never a set_fmt(TRY) at the sink pad.
>>
>> But perhaps the following should be added to this patch as well. It
>> makes the
>> assumption that the TRY code at the sink pad is the same as the default
>> active code set from csi_registered().
>>
>> Or maybe I should ask the question, what should drivers do in set_fmt(TRY)
>> at their source pads, if there was no prior set_fmt(TRY) at their sink pads?
> Drivers can set the initial TRY value by implementing the init_cfg pad op.
> See e.g. drivers/media/platform/vimc/vimc-sensor.c.

Thanks. I will send a patch that implements init_cfg pad op in all
imx-media subdevs.

Steve

>
> I think you have two choices: either set it to some default format, or copy
> the current ACTIVE format. It doesn't really matter, as long as there is
> something valid.
>
> Many drivers will likely fail this test.
>
> Regards,
>
> 	Hans
>
>
  
Steve Longerbeam Feb. 10, 2018, 1:43 a.m. UTC | #4
Hi Philipp,


On 02/08/2018 12:01 PM, Hans Verkuil wrote:
> On 02/08/2018 08:27 PM, Steve Longerbeam wrote:
>>
>> On 02/08/2018 06:47 AM, Philipp Zabel wrote:
>>> If no imx_media_pixfmt is found for a given mbus format code,
>>> we shouldn't crash. Return -EINVAL for any index.
>> Hi Philipp,
>>
>> If imx_media_find_mbus_format() returns NULL at that location, it means
>> the current format has an invalid pixel code. It's not possible for an
>> ACTIVE
>> format to have an invalid code, so it must be a TRY format with an
>> uninitialized
>> (zero) code. Which makes sense if enum_mbus_code(TRY) pad op is called
>> before
>> set_fmt(TRY), at the sink pad, was *ever* called. Am I right?
>>
>> It looks there is another location where this could possibly happen, in
>> csi_try_fmt().
>> In that case it would happen if set_fmt(TRY) is called on a source pad,
>> before
>> set_fmt(TRY) was ever called at the sink pad. That's a weird corner case
>> because
>> it's not clear what a set_fmt(TRY) at the source pads should choose for
>> pixel
>> code if there was never a set_fmt(TRY) at the sink pad.
>>
>> But perhaps the following should be added to this patch as well. It
>> makes the
>> assumption that the TRY code at the sink pad is the same as the default
>> active code set from csi_registered().
>>
>> Or maybe I should ask the question, what should drivers do in set_fmt(TRY)
>> at their source pads, if there was no prior set_fmt(TRY) at their sink pads?
> Drivers can set the initial TRY value by implementing the init_cfg pad op.
> See e.g. drivers/media/platform/vimc/vimc-sensor.c.

I *think* by implementing init_cfg in the CSI, it will prevent the
NULL deref in csi_enum_mbus_code(). However I think this patch
is a good idea in any case.

Steve

>
>>> Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
>>> ---
>>>    drivers/staging/media/imx/imx-media-csi.c | 4 ++++
>>>    1 file changed, 4 insertions(+)
>>>
>>> diff --git a/drivers/staging/media/imx/imx-media-csi.c b/drivers/staging/media/imx/imx-media-csi.c
>>> index eb7be5093a9d..89903f267d60 100644
>>> --- a/drivers/staging/media/imx/imx-media-csi.c
>>> +++ b/drivers/staging/media/imx/imx-media-csi.c
>>> @@ -1138,6 +1138,10 @@ static int csi_enum_mbus_code(struct v4l2_subdev *sd,
>>>    
>>>    	infmt = __csi_get_fmt(priv, cfg, CSI_SINK_PAD, code->which);
>>>    	incc = imx_media_find_mbus_format(infmt->code, CS_SEL_ANY, true);
>>> +	if (!incc) {
>>> +		ret = -EINVAL;
>>> +		goto out;
>>> +	}
>>>    
>>>    	switch (code->pad) {
>>>    	case CSI_SINK_PAD:
  
Philipp Zabel Nov. 5, 2018, 2:59 p.m. UTC | #5
Hi Steve,

On Fri, 2018-02-09 at 17:43 -0800, Steve Longerbeam wrote:
[...]
> I *think* by implementing init_cfg in the CSI, it will prevent the
> NULL deref in csi_enum_mbus_code(). However I think this patch
> is a good idea in any case.

Ack on both. Can we still get this patch applied?

regards
Philipp
  

Patch

diff --git a/drivers/staging/media/imx/imx-media-csi.c b/drivers/staging/media/imx/imx-media-csi.c
index eb7be5093a9d..89903f267d60 100644
--- a/drivers/staging/media/imx/imx-media-csi.c
+++ b/drivers/staging/media/imx/imx-media-csi.c
@@ -1138,6 +1138,10 @@  static int csi_enum_mbus_code(struct v4l2_subdev *sd,
 
 	infmt = __csi_get_fmt(priv, cfg, CSI_SINK_PAD, code->which);
 	incc = imx_media_find_mbus_format(infmt->code, CS_SEL_ANY, true);
+	if (!incc) {
+		ret = -EINVAL;
+		goto out;
+	}
 
 	switch (code->pad) {
 	case CSI_SINK_PAD: