[PATCH] [media] v4l2-async: Always unregister the subdev on failure

[PATCH] [media] v4l2-async: Always unregister the subdev on failure

[Alban Bedel]

In v4l2_async_test_notify() if the registered_async callback or the
complete notifier returns an error the subdev is not unregistered.
This leave paths where v4l2_async_register_subdev() can fail but
leave the subdev still registered.

Add the required calls to v4l2_device_unregister_subdev() to plug
these holes.

Signed-off-by: Alban Bedel <alban.bedel@avionic-design.de>
---
 drivers/media/v4l2-core/v4l2-async.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/media/v4l2-core/v4l2-async.c b/drivers/media/v4l2-core/v4l2-async.c
index ceb28d4..43393f8 100644
--- a/drivers/media/v4l2-core/v4l2-async.c
+++ b/drivers/media/v4l2-core/v4l2-async.c
@@ -121,13 +121,19 @@ static int v4l2_async_test_notify(struct v4l2_async_notifier *notifier,
 
 	ret = v4l2_subdev_call(sd, core, registered_async);
 	if (ret < 0 && ret != -ENOIOCTLCMD) {
+		v4l2_device_unregister_subdev(sd);
 		if (notifier->unbind)
 			notifier->unbind(notifier, sd, asd);
 		return ret;
 	}
 
-	if (list_empty(&notifier->waiting) && notifier->complete)
-		return notifier->complete(notifier);
+	if (list_empty(&notifier->waiting) && notifier->complete) {
+		ret = notifier->complete(notifier);
+		if (ret < 0) {
+			v4l2_device_unregister_subdev(sd);
+			return ret;
+		}
+	}
 
 	return 0;
 }
-- 
2.8.2

Re: [PATCH] [media] v4l2-async: Always unregister the subdev on failure

[Javier Martinez Canillas]

Hello Alban,

On 05/11/2016 11:40 AM, Alban Bedel wrote:
> In v4l2_async_test_notify() if the registered_async callback or the
> complete notifier returns an error the subdev is not unregistered.
> This leave paths where v4l2_async_register_subdev() can fail but
> leave the subdev still registered.
> 
> Add the required calls to v4l2_device_unregister_subdev() to plug
> these holes.
> 
> Signed-off-by: Alban Bedel <alban.bedel@avionic-design.de>
> ---
>  drivers/media/v4l2-core/v4l2-async.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/media/v4l2-core/v4l2-async.c b/drivers/media/v4l2-core/v4l2-async.c
> index ceb28d4..43393f8 100644
> --- a/drivers/media/v4l2-core/v4l2-async.c
> +++ b/drivers/media/v4l2-core/v4l2-async.c
> @@ -121,13 +121,19 @@ static int v4l2_async_test_notify(struct v4l2_async_notifier *notifier,
>  
>  	ret = v4l2_subdev_call(sd, core, registered_async);
>  	if (ret < 0 && ret != -ENOIOCTLCMD) {
> +		v4l2_device_unregister_subdev(sd);
>  		if (notifier->unbind)
>  			notifier->unbind(notifier, sd, asd);
>  		return ret;
>  	}
>  
> -	if (list_empty(&notifier->waiting) && notifier->complete)
> -		return notifier->complete(notifier);
> +	if (list_empty(&notifier->waiting) && notifier->complete) {
> +		ret = notifier->complete(notifier);
> +		if (ret < 0) {
> +			v4l2_device_unregister_subdev(sd);

Isn't a call to notifier->unbind() missing here as well?

Also, I think the error path is becoming too duplicated and complex, so
maybe we can have a single error path and use goto labels as is common
in Linux? For example something like the following (not tested) can be
squashed on top of your change:

diff --git a/drivers/media/v4l2-core/v4l2-async.c b/drivers/media/v4l2-core/v4l2-async.c
index 43393f8c1312..abe512d0b4cb 100644
--- a/drivers/media/v4l2-core/v4l2-async.c
+++ b/drivers/media/v4l2-core/v4l2-async.c
@@ -113,29 +113,28 @@ static int v4l2_async_test_notify(struct v4l2_async_notifier *notifier,
 	list_move(&sd->async_list, &notifier->done);
 
 	ret = v4l2_device_register_subdev(notifier->v4l2_dev, sd);
-	if (ret < 0) {
-		if (notifier->unbind)
-			notifier->unbind(notifier, sd, asd);
-		return ret;
-	}
+	if (ret < 0)
+		goto err_subdev_register;
 
 	ret = v4l2_subdev_call(sd, core, registered_async);
-	if (ret < 0 && ret != -ENOIOCTLCMD) {
-		v4l2_device_unregister_subdev(sd);
-		if (notifier->unbind)
-			notifier->unbind(notifier, sd, asd);
-		return ret;
-	}
+	if (ret < 0 && ret != -ENOIOCTLCMD)
+		goto err_subdev_call;
 
 	if (list_empty(&notifier->waiting) && notifier->complete) {
 		ret = notifier->complete(notifier);
-		if (ret < 0) {
-			v4l2_device_unregister_subdev(sd);
-			return ret;
-		}
+		if (ret < 0)
+			goto err_subdev_call;
 	}
 
 	return 0;
+
+err_subdev_call:
+	v4l2_device_unregister_subdev(sd);
+err_subdev_register:
+	if (notifier->unbind)
+		notifier->unbind(notifier, sd, asd);
+
+	return ret;
 }
 
 static void v4l2_async_cleanup(struct v4l2_subdev *sd)

Best regards,
-- 
Javier Martinez Canillas
Open Source Group
Samsung Research America

Re: [PATCH] [media] v4l2-async: Always unregister the subdev on failure

[Alban Bedel]

[-- Attachment #1: Type: text/plain, Size: 1983 bytes --]

On Wed, 11 May 2016 12:22:44 -0400
Javier Martinez Canillas <javier@osg.samsung.com> wrote:

> Hello Alban,
> 
> On 05/11/2016 11:40 AM, Alban Bedel wrote:
> > In v4l2_async_test_notify() if the registered_async callback or the
> > complete notifier returns an error the subdev is not unregistered.
> > This leave paths where v4l2_async_register_subdev() can fail but
> > leave the subdev still registered.
> > 
> > Add the required calls to v4l2_device_unregister_subdev() to plug
> > these holes.
> > 
> > Signed-off-by: Alban Bedel <alban.bedel@avionic-design.de>
> > ---
> >  drivers/media/v4l2-core/v4l2-async.c | 10 ++++++++--
> >  1 file changed, 8 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/media/v4l2-core/v4l2-async.c b/drivers/media/v4l2-core/v4l2-async.c
> > index ceb28d4..43393f8 100644
> > --- a/drivers/media/v4l2-core/v4l2-async.c
> > +++ b/drivers/media/v4l2-core/v4l2-async.c
> > @@ -121,13 +121,19 @@ static int v4l2_async_test_notify(struct v4l2_async_notifier *notifier,
> >  
> >  	ret = v4l2_subdev_call(sd, core, registered_async);
> >  	if (ret < 0 && ret != -ENOIOCTLCMD) {
> > +		v4l2_device_unregister_subdev(sd);
> >  		if (notifier->unbind)
> >  			notifier->unbind(notifier, sd, asd);
> >  		return ret;
> >  	}
> >  
> > -	if (list_empty(&notifier->waiting) && notifier->complete)
> > -		return notifier->complete(notifier);
> > +	if (list_empty(&notifier->waiting) && notifier->complete) {
> > +		ret = notifier->complete(notifier);
> > +		if (ret < 0) {
> > +			v4l2_device_unregister_subdev(sd);
> 
> Isn't a call to notifier->unbind() missing here as well?
> 
> Also, I think the error path is becoming too duplicated and complex, so
> maybe we can have a single error path and use goto labels as is common
> in Linux? For example something like the following (not tested) can be
> squashed on top of your change:

Yes, that look better. I'll test it and report tomorrow.

Alban

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH] [media] v4l2-async: Always unregister the subdev on failure

[Hans Verkuil]

On 05/11/2016 06:32 PM, Alban Bedel wrote:
> On Wed, 11 May 2016 12:22:44 -0400
> Javier Martinez Canillas <javier@osg.samsung.com> wrote:
> 
>> Hello Alban,
>>
>> On 05/11/2016 11:40 AM, Alban Bedel wrote:
>>> In v4l2_async_test_notify() if the registered_async callback or the
>>> complete notifier returns an error the subdev is not unregistered.
>>> This leave paths where v4l2_async_register_subdev() can fail but
>>> leave the subdev still registered.
>>>
>>> Add the required calls to v4l2_device_unregister_subdev() to plug
>>> these holes.
>>>
>>> Signed-off-by: Alban Bedel <alban.bedel@avionic-design.de>
>>> ---
>>>  drivers/media/v4l2-core/v4l2-async.c | 10 ++++++++--
>>>  1 file changed, 8 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/media/v4l2-core/v4l2-async.c b/drivers/media/v4l2-core/v4l2-async.c
>>> index ceb28d4..43393f8 100644
>>> --- a/drivers/media/v4l2-core/v4l2-async.c
>>> +++ b/drivers/media/v4l2-core/v4l2-async.c
>>> @@ -121,13 +121,19 @@ static int v4l2_async_test_notify(struct v4l2_async_notifier *notifier,
>>>  
>>>  	ret = v4l2_subdev_call(sd, core, registered_async);
>>>  	if (ret < 0 && ret != -ENOIOCTLCMD) {
>>> +		v4l2_device_unregister_subdev(sd);
>>>  		if (notifier->unbind)
>>>  			notifier->unbind(notifier, sd, asd);
>>>  		return ret;
>>>  	}
>>>  
>>> -	if (list_empty(&notifier->waiting) && notifier->complete)
>>> -		return notifier->complete(notifier);
>>> +	if (list_empty(&notifier->waiting) && notifier->complete) {
>>> +		ret = notifier->complete(notifier);
>>> +		if (ret < 0) {
>>> +			v4l2_device_unregister_subdev(sd);
>>
>> Isn't a call to notifier->unbind() missing here as well?
>>
>> Also, I think the error path is becoming too duplicated and complex, so
>> maybe we can have a single error path and use goto labels as is common
>> in Linux? For example something like the following (not tested) can be
>> squashed on top of your change:
> 
> Yes, that look better. I'll test it and report tomorrow.

I haven't heard anything back about this. Did you manage to test it?

Regards,

	Hans

Re: [PATCH] [media] v4l2-async: Always unregister the subdev on failure

[Alban Bedel]

[-- Attachment #1: Type: text/plain, Size: 2382 bytes --]

On Fri, 1 Jul 2016 13:55:44 +0200
Hans Verkuil <hverkuil@xs4all.nl> wrote:

> On 05/11/2016 06:32 PM, Alban Bedel wrote:
> > On Wed, 11 May 2016 12:22:44 -0400
> > Javier Martinez Canillas <javier@osg.samsung.com> wrote:
> >   
> >> Hello Alban,
> >>
> >> On 05/11/2016 11:40 AM, Alban Bedel wrote:  
> >>> In v4l2_async_test_notify() if the registered_async callback or the
> >>> complete notifier returns an error the subdev is not unregistered.
> >>> This leave paths where v4l2_async_register_subdev() can fail but
> >>> leave the subdev still registered.
> >>>
> >>> Add the required calls to v4l2_device_unregister_subdev() to plug
> >>> these holes.
> >>>
> >>> Signed-off-by: Alban Bedel <alban.bedel@avionic-design.de>
> >>> ---
> >>>  drivers/media/v4l2-core/v4l2-async.c | 10 ++++++++--
> >>>  1 file changed, 8 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/media/v4l2-core/v4l2-async.c b/drivers/media/v4l2-core/v4l2-async.c
> >>> index ceb28d4..43393f8 100644
> >>> --- a/drivers/media/v4l2-core/v4l2-async.c
> >>> +++ b/drivers/media/v4l2-core/v4l2-async.c
> >>> @@ -121,13 +121,19 @@ static int v4l2_async_test_notify(struct v4l2_async_notifier *notifier,
> >>>  
> >>>  	ret = v4l2_subdev_call(sd, core, registered_async);
> >>>  	if (ret < 0 && ret != -ENOIOCTLCMD) {
> >>> +		v4l2_device_unregister_subdev(sd);
> >>>  		if (notifier->unbind)
> >>>  			notifier->unbind(notifier, sd, asd);
> >>>  		return ret;
> >>>  	}
> >>>  
> >>> -	if (list_empty(&notifier->waiting) && notifier->complete)
> >>> -		return notifier->complete(notifier);
> >>> +	if (list_empty(&notifier->waiting) && notifier->complete) {
> >>> +		ret = notifier->complete(notifier);
> >>> +		if (ret < 0) {
> >>> +			v4l2_device_unregister_subdev(sd);  
> >>
> >> Isn't a call to notifier->unbind() missing here as well?
> >>
> >> Also, I think the error path is becoming too duplicated and complex, so
> >> maybe we can have a single error path and use goto labels as is common
> >> in Linux? For example something like the following (not tested) can be
> >> squashed on top of your change:  
> > 
> > Yes, that look better. I'll test it and report tomorrow.  
> 
> I haven't heard anything back about this. Did you manage to test it?

Yes, that's working fine. Sorry for the delay, I'm sending the v2 patch.

Alban


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]