diff mbox

v4l: rcar-vin: Implement V4L2 video node release handler

Message ID 20171115224907.392-1-laurent.pinchart+renesas@ideasonboard.com (mailing list archive)
State Not Applicable
Delegated to: Geert Uytterhoeven
Headers show

Commit Message

Laurent Pinchart Nov. 15, 2017, 10:49 p.m. UTC
The rvin_dev data structure contains driver private data for an instance
of the VIN. It is allocated dynamically at probe time, and must be freed
once all users are gone.

The structure is currently allocated with devm_kzalloc(), which results
in memory being freed when the device is unbound. If a userspace
application is currently performing an ioctl call, or just keeps the
device node open and closes it later, this will lead to accessing freed
memory.

Fix the problem by implementing a V4L2 release handler for the video
node associated with the VIN instance (called when the last user of the
video node releases it), and freeing memory explicitly from the release
handler.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
---
 drivers/media/platform/rcar-vin/rcar-core.c | 29 +++++++++++++++++++----------
 drivers/media/platform/rcar-vin/rcar-v4l2.c |  9 ++++++++-
 2 files changed, 27 insertions(+), 11 deletions(-)

Comments

Niklas Söderlund Nov. 15, 2017, 11:34 p.m. UTC | #1
Hi Laurent,

Thanks for your patch.

On 2017-11-16 00:49:07 +0200, Laurent Pinchart wrote:
> The rvin_dev data structure contains driver private data for an instance
> of the VIN. It is allocated dynamically at probe time, and must be freed
> once all users are gone.
> 
> The structure is currently allocated with devm_kzalloc(), which results
> in memory being freed when the device is unbound. If a userspace
> application is currently performing an ioctl call, or just keeps the
> device node open and closes it later, this will lead to accessing freed
> memory.
> 
> Fix the problem by implementing a V4L2 release handler for the video
> node associated with the VIN instance (called when the last user of the
> video node releases it), and freeing memory explicitly from the release
> handler.
> 
> Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Acked-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>

This patch is based on-top of the VIN Gen3 enablement series not yet 
upstream. This is OK for me, just wanted to check that this was the 
intention as to minimize conflicts between the two.

> ---
>  drivers/media/platform/rcar-vin/rcar-core.c | 29 +++++++++++++++++++----------
>  drivers/media/platform/rcar-vin/rcar-v4l2.c |  9 ++++++++-
>  2 files changed, 27 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/media/platform/rcar-vin/rcar-core.c b/drivers/media/platform/rcar-vin/rcar-core.c
> index 495610949457..bd7976efa1fb 100644
> --- a/drivers/media/platform/rcar-vin/rcar-core.c
> +++ b/drivers/media/platform/rcar-vin/rcar-core.c
> @@ -1208,7 +1208,7 @@ static int rcar_vin_probe(struct platform_device *pdev)
>  	struct resource *mem;
>  	int irq, ret;
>  
> -	vin = devm_kzalloc(&pdev->dev, sizeof(*vin), GFP_KERNEL);
> +	vin = kzalloc(sizeof(*vin), GFP_KERNEL);
>  	if (!vin)
>  		return -ENOMEM;
>  
> @@ -1224,20 +1224,26 @@ static int rcar_vin_probe(struct platform_device *pdev)
>  		vin->info = attr->data;
>  
>  	mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> -	if (mem == NULL)
> -		return -EINVAL;
> +	if (mem == NULL) {
> +		ret = -EINVAL;
> +		goto error_free;
> +	}
>  
>  	vin->base = devm_ioremap_resource(vin->dev, mem);
> -	if (IS_ERR(vin->base))
> -		return PTR_ERR(vin->base);
> +	if (IS_ERR(vin->base)) {
> +		ret = PTR_ERR(vin->base);
> +		goto error_free;
> +	}
>  
>  	irq = platform_get_irq(pdev, 0);
> -	if (irq < 0)
> -		return irq;
> +	if (irq < 0) {
> +		ret = irq;
> +		goto error_free;
> +	}
>  
>  	ret = rvin_dma_probe(vin, irq);
>  	if (ret)
> -		return ret;
> +		goto error_free;
>  
>  	platform_set_drvdata(pdev, vin);
>  	if (vin->info->use_mc)
> @@ -1245,15 +1251,18 @@ static int rcar_vin_probe(struct platform_device *pdev)
>  	else
>  		ret = rvin_digital_graph_init(vin);
>  	if (ret < 0)
> -		goto error;
> +		goto error_dma;
>  
>  	pm_suspend_ignore_children(&pdev->dev, true);
>  	pm_runtime_enable(&pdev->dev);
>  
>  	return 0;
> -error:
> +
> +error_dma:
>  	rvin_dma_remove(vin);
>  	v4l2_async_notifier_cleanup(&vin->notifier);
> +error_free:
> +	kfree(vin);
>  
>  	return ret;
>  }
> diff --git a/drivers/media/platform/rcar-vin/rcar-v4l2.c b/drivers/media/platform/rcar-vin/rcar-v4l2.c
> index 2c14d44950b2..25f1d24c1d2d 100644
> --- a/drivers/media/platform/rcar-vin/rcar-v4l2.c
> +++ b/drivers/media/platform/rcar-vin/rcar-v4l2.c
> @@ -1026,6 +1026,13 @@ static void rvin_notify(struct v4l2_subdev *sd,
>  	}
>  }
>  
> +static void rvin_v4l2_release(struct video_device *vdev)
> +{
> +	struct rvin_dev *vin = container_of(vdev, struct rvin_dev, vdev);
> +
> +	kfree(vin);
> +}
> +
>  int rvin_v4l2_register(struct rvin_dev *vin)
>  {
>  	struct video_device *vdev = &vin->vdev;
> @@ -1038,7 +1045,7 @@ int rvin_v4l2_register(struct rvin_dev *vin)
>  	vdev->queue = &vin->queue;
>  	snprintf(vdev->name, sizeof(vdev->name), "%s %s", KBUILD_MODNAME,
>  		 dev_name(vin->dev));
> -	vdev->release = video_device_release_empty;
> +	vdev->release = rvin_v4l2_release;
>  	vdev->lock = &vin->lock;
>  	vdev->device_caps = V4L2_CAP_VIDEO_CAPTURE | V4L2_CAP_STREAMING |
>  		V4L2_CAP_READWRITE;
> -- 
> Regards,
> 
> Laurent Pinchart
>
Laurent Pinchart Nov. 16, 2017, 12:27 a.m. UTC | #2
Hi Niklas,

On Thursday, 16 November 2017 01:34:33 EET Niklas Söderlund wrote:
> On 2017-11-16 00:49:07 +0200, Laurent Pinchart wrote:
> > The rvin_dev data structure contains driver private data for an instance
> > of the VIN. It is allocated dynamically at probe time, and must be freed
> > once all users are gone.
> > 
> > The structure is currently allocated with devm_kzalloc(), which results
> > in memory being freed when the device is unbound. If a userspace
> > application is currently performing an ioctl call, or just keeps the
> > device node open and closes it later, this will lead to accessing freed
> > memory.
> > 
> > Fix the problem by implementing a V4L2 release handler for the video
> > node associated with the VIN instance (called when the last user of the
> > video node releases it), and freeing memory explicitly from the release
> > handler.
> > 
> > Signed-off-by: Laurent Pinchart
> > <laurent.pinchart+renesas@ideasonboard.com>
> 
> Acked-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
> 
> This patch is based on-top of the VIN Gen3 enablement series not yet
> upstream. This is OK for me, just wanted to check that this was the
> intention as to minimize conflicts between the two.

Yes, that's my intention. The patch should be included, or possibly squashed 
in, your development branch.
Hans Verkuil Dec. 4, 2017, 9:05 a.m. UTC | #3
Hi Niklas,

On 11/16/2017 01:27 AM, Laurent Pinchart wrote:
> Hi Niklas,
> 
> On Thursday, 16 November 2017 01:34:33 EET Niklas Söderlund wrote:
>> On 2017-11-16 00:49:07 +0200, Laurent Pinchart wrote:
>>> The rvin_dev data structure contains driver private data for an instance
>>> of the VIN. It is allocated dynamically at probe time, and must be freed
>>> once all users are gone.
>>>
>>> The structure is currently allocated with devm_kzalloc(), which results
>>> in memory being freed when the device is unbound. If a userspace
>>> application is currently performing an ioctl call, or just keeps the
>>> device node open and closes it later, this will lead to accessing freed
>>> memory.
>>>
>>> Fix the problem by implementing a V4L2 release handler for the video
>>> node associated with the VIN instance (called when the last user of the
>>> video node releases it), and freeing memory explicitly from the release
>>> handler.
>>>
>>> Signed-off-by: Laurent Pinchart
>>> <laurent.pinchart+renesas@ideasonboard.com>
>>
>> Acked-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
>>
>> This patch is based on-top of the VIN Gen3 enablement series not yet
>> upstream. This is OK for me, just wanted to check that this was the
>> intention as to minimize conflicts between the two.
> 
> Yes, that's my intention. The patch should be included, or possibly squashed 
> in, your development branch.
> 

Has this patch been added in your v8 series? If not, can you add it when you
post a v9?

Thanks,

	Hans
Niklas Söderlund Dec. 4, 2017, 1:34 p.m. UTC | #4
Hi Hans,

On 2017-12-04 10:05:35 +0100, Hans Verkuil wrote:
> Hi Niklas,
> 
> On 11/16/2017 01:27 AM, Laurent Pinchart wrote:
> > Hi Niklas,
> > 
> > On Thursday, 16 November 2017 01:34:33 EET Niklas Söderlund wrote:
> >> On 2017-11-16 00:49:07 +0200, Laurent Pinchart wrote:
> >>> The rvin_dev data structure contains driver private data for an instance
> >>> of the VIN. It is allocated dynamically at probe time, and must be freed
> >>> once all users are gone.
> >>>
> >>> The structure is currently allocated with devm_kzalloc(), which results
> >>> in memory being freed when the device is unbound. If a userspace
> >>> application is currently performing an ioctl call, or just keeps the
> >>> device node open and closes it later, this will lead to accessing freed
> >>> memory.
> >>>
> >>> Fix the problem by implementing a V4L2 release handler for the video
> >>> node associated with the VIN instance (called when the last user of the
> >>> video node releases it), and freeing memory explicitly from the release
> >>> handler.
> >>>
> >>> Signed-off-by: Laurent Pinchart
> >>> <laurent.pinchart+renesas@ideasonboard.com>
> >>
> >> Acked-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
> >>
> >> This patch is based on-top of the VIN Gen3 enablement series not yet
> >> upstream. This is OK for me, just wanted to check that this was the
> >> intention as to minimize conflicts between the two.
> > 
> > Yes, that's my intention. The patch should be included, or possibly squashed 
> > in, your development branch.
> > 
> 
> Has this patch been added in your v8 series? If not, can you add it when you
> post a v9?

This patch needs more work, with the video device now registered at 
complete() time and unregistered at unbind() time. Applying this would 
free the rcar-vin private data structure at unbind() which probably not 
what we want.

I think this issue should be addressed but maybe together with a 
patch-set targeting the generic problem with video device lifetimes in 
v4l2 framework? For now I would be happy to focus on getting Gen3 
support picked-up and observe what Laurent's work on lifetime issues 
brings and adept the rcar-vin driver to take advantage of that once it's 
ready.

> 
> Thanks,
> 
> 	Hans
Hans Verkuil Dec. 4, 2017, 1:41 p.m. UTC | #5
On 12/04/2017 02:34 PM, Niklas Söderlund wrote:
> Hi Hans,
> 
> On 2017-12-04 10:05:35 +0100, Hans Verkuil wrote:
>> Hi Niklas,
>>
>> On 11/16/2017 01:27 AM, Laurent Pinchart wrote:
>>> Hi Niklas,
>>>
>>> On Thursday, 16 November 2017 01:34:33 EET Niklas Söderlund wrote:
>>>> On 2017-11-16 00:49:07 +0200, Laurent Pinchart wrote:
>>>>> The rvin_dev data structure contains driver private data for an instance
>>>>> of the VIN. It is allocated dynamically at probe time, and must be freed
>>>>> once all users are gone.
>>>>>
>>>>> The structure is currently allocated with devm_kzalloc(), which results
>>>>> in memory being freed when the device is unbound. If a userspace
>>>>> application is currently performing an ioctl call, or just keeps the
>>>>> device node open and closes it later, this will lead to accessing freed
>>>>> memory.
>>>>>
>>>>> Fix the problem by implementing a V4L2 release handler for the video
>>>>> node associated with the VIN instance (called when the last user of the
>>>>> video node releases it), and freeing memory explicitly from the release
>>>>> handler.
>>>>>
>>>>> Signed-off-by: Laurent Pinchart
>>>>> <laurent.pinchart+renesas@ideasonboard.com>
>>>>
>>>> Acked-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
>>>>
>>>> This patch is based on-top of the VIN Gen3 enablement series not yet
>>>> upstream. This is OK for me, just wanted to check that this was the
>>>> intention as to minimize conflicts between the two.
>>>
>>> Yes, that's my intention. The patch should be included, or possibly squashed 
>>> in, your development branch.
>>>
>>
>> Has this patch been added in your v8 series? If not, can you add it when you
>> post a v9?
> 
> This patch needs more work, with the video device now registered at 
> complete() time and unregistered at unbind() time. Applying this would 
> free the rcar-vin private data structure at unbind() which probably not 
> what we want.
> 
> I think this issue should be addressed but maybe together with a 
> patch-set targeting the generic problem with video device lifetimes in 
> v4l2 framework? For now I would be happy to focus on getting Gen3 
> support picked-up and observe what Laurent's work on lifetime issues 
> brings and adept the rcar-vin driver to take advantage of that once it's 
> ready.

OK. I marked the patch as "Obsoleted" so it doesn't stick around in my patch list.

Regards,

	Hans
diff mbox

Patch

diff --git a/drivers/media/platform/rcar-vin/rcar-core.c b/drivers/media/platform/rcar-vin/rcar-core.c
index 495610949457..bd7976efa1fb 100644
--- a/drivers/media/platform/rcar-vin/rcar-core.c
+++ b/drivers/media/platform/rcar-vin/rcar-core.c
@@ -1208,7 +1208,7 @@  static int rcar_vin_probe(struct platform_device *pdev)
 	struct resource *mem;
 	int irq, ret;
 
-	vin = devm_kzalloc(&pdev->dev, sizeof(*vin), GFP_KERNEL);
+	vin = kzalloc(sizeof(*vin), GFP_KERNEL);
 	if (!vin)
 		return -ENOMEM;
 
@@ -1224,20 +1224,26 @@  static int rcar_vin_probe(struct platform_device *pdev)
 		vin->info = attr->data;
 
 	mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
-	if (mem == NULL)
-		return -EINVAL;
+	if (mem == NULL) {
+		ret = -EINVAL;
+		goto error_free;
+	}
 
 	vin->base = devm_ioremap_resource(vin->dev, mem);
-	if (IS_ERR(vin->base))
-		return PTR_ERR(vin->base);
+	if (IS_ERR(vin->base)) {
+		ret = PTR_ERR(vin->base);
+		goto error_free;
+	}
 
 	irq = platform_get_irq(pdev, 0);
-	if (irq < 0)
-		return irq;
+	if (irq < 0) {
+		ret = irq;
+		goto error_free;
+	}
 
 	ret = rvin_dma_probe(vin, irq);
 	if (ret)
-		return ret;
+		goto error_free;
 
 	platform_set_drvdata(pdev, vin);
 	if (vin->info->use_mc)
@@ -1245,15 +1251,18 @@  static int rcar_vin_probe(struct platform_device *pdev)
 	else
 		ret = rvin_digital_graph_init(vin);
 	if (ret < 0)
-		goto error;
+		goto error_dma;
 
 	pm_suspend_ignore_children(&pdev->dev, true);
 	pm_runtime_enable(&pdev->dev);
 
 	return 0;
-error:
+
+error_dma:
 	rvin_dma_remove(vin);
 	v4l2_async_notifier_cleanup(&vin->notifier);
+error_free:
+	kfree(vin);
 
 	return ret;
 }
diff --git a/drivers/media/platform/rcar-vin/rcar-v4l2.c b/drivers/media/platform/rcar-vin/rcar-v4l2.c
index 2c14d44950b2..25f1d24c1d2d 100644
--- a/drivers/media/platform/rcar-vin/rcar-v4l2.c
+++ b/drivers/media/platform/rcar-vin/rcar-v4l2.c
@@ -1026,6 +1026,13 @@  static void rvin_notify(struct v4l2_subdev *sd,
 	}
 }
 
+static void rvin_v4l2_release(struct video_device *vdev)
+{
+	struct rvin_dev *vin = container_of(vdev, struct rvin_dev, vdev);
+
+	kfree(vin);
+}
+
 int rvin_v4l2_register(struct rvin_dev *vin)
 {
 	struct video_device *vdev = &vin->vdev;
@@ -1038,7 +1045,7 @@  int rvin_v4l2_register(struct rvin_dev *vin)
 	vdev->queue = &vin->queue;
 	snprintf(vdev->name, sizeof(vdev->name), "%s %s", KBUILD_MODNAME,
 		 dev_name(vin->dev));
-	vdev->release = video_device_release_empty;
+	vdev->release = rvin_v4l2_release;
 	vdev->lock = &vin->lock;
 	vdev->device_caps = V4L2_CAP_VIDEO_CAPTURE | V4L2_CAP_STREAMING |
 		V4L2_CAP_READWRITE;