[PATCH] dvb-core: Fix potential mutex_unlock without mutex_lock in dvb_dvr_read

[PATCH] dvb-core: Fix potential mutex_unlock without mutex_lock in dvb_dvr_read

[Simon Arlott]

dvb_dvr_read may unlock the dmxdev mutex and return -ENODEV,
except this function is a file op and will never be called
with the mutex held.

There's existing mutex_lock and mutex_unlock around the actual
read but it's commented out. These should probably be uncommented
but the read blocks and this could block another non-blocking
reader on the mutex instead.

This change comments out the extra mutex_unlock.

Signed-off-by: Simon Arlott <simon@fire.lp0.eu>
---
This has been on my TODO list for far too long... I did come
up with a mutex_trylock/mutex_lock_interruptible version but
claiming that it'll block when it may not doesn't make sense
(and any blocking read would cause all non-blocking reads to
continually return -EWOULDBLOCK until there is data).

 drivers/media/dvb/dvb-core/dmxdev.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/media/dvb/dvb-core/dmxdev.c b/drivers/media/dvb/dvb-core/dmxdev.c
index c35fbb8..d6d098a 100644
--- a/drivers/media/dvb/dvb-core/dmxdev.c
+++ b/drivers/media/dvb/dvb-core/dmxdev.c
@@ -247,7 +247,7 @@ static ssize_t dvb_dvr_read(struct file *file, char __user *buf, size_t count,
 	int ret;
 
 	if (dmxdev->exit) {
-		mutex_unlock(&dmxdev->mutex);
+		//mutex_unlock(&dmxdev->mutex);
 		return -ENODEV;
 	}
 
-- 
1.6.2.2

-- 
Simon Arlott

Re: [PATCH] dvb-core: Fix potential mutex_unlock without mutex_lock in dvb_dvr_read

[Andrew Morton]

On Thu, 23 Apr 2009 18:32:13 +0100
Simon Arlott <simon@fire.lp0.eu> wrote:

> dvb_dvr_read may unlock the dmxdev mutex and return -ENODEV,
> except this function is a file op and will never be called
> with the mutex held.
> 
> There's existing mutex_lock and mutex_unlock around the actual
> read but it's commented out. These should probably be uncommented
> but the read blocks and this could block another non-blocking
> reader on the mutex instead.
> 
> This change comments out the extra mutex_unlock.
> 
> Signed-off-by: Simon Arlott <simon@fire.lp0.eu>
> ---
> This has been on my TODO list for far too long... I did come
> up with a mutex_trylock/mutex_lock_interruptible version but
> claiming that it'll block when it may not doesn't make sense
> (and any blocking read would cause all non-blocking reads to
> continually return -EWOULDBLOCK until there is data).
> 
>  drivers/media/dvb/dvb-core/dmxdev.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/drivers/media/dvb/dvb-core/dmxdev.c b/drivers/media/dvb/dvb-core/dmxdev.c
> index c35fbb8..d6d098a 100644
> --- a/drivers/media/dvb/dvb-core/dmxdev.c
> +++ b/drivers/media/dvb/dvb-core/dmxdev.c
> @@ -247,7 +247,7 @@ static ssize_t dvb_dvr_read(struct file *file, char __user *buf, size_t count,
>  	int ret;
>  
>  	if (dmxdev->exit) {
> -		mutex_unlock(&dmxdev->mutex);
> +		//mutex_unlock(&dmxdev->mutex);
>  		return -ENODEV;
>  	}

Is there any value in retaining all the commented-out lock operations,
or can we zap 'em?

--- a/drivers/media/dvb/dvb-core/dmxdev.c~dvb-core-fix-potential-mutex_unlock-without-mutex_lock-in-dvb_dvr_read-fix
+++ a/drivers/media/dvb/dvb-core/dmxdev.c
@@ -244,19 +244,13 @@ static ssize_t dvb_dvr_read(struct file 
 {
 	struct dvb_device *dvbdev = file->private_data;
 	struct dmxdev *dmxdev = dvbdev->priv;
-	int ret;
 
-	if (dmxdev->exit) {
-		//mutex_unlock(&dmxdev->mutex);
+	if (dmxdev->exit)
 		return -ENODEV;
-	}
 
-	//mutex_lock(&dmxdev->mutex);
-	ret = dvb_dmxdev_buffer_read(&dmxdev->dvr_buffer,
-				     file->f_flags & O_NONBLOCK,
-				     buf, count, ppos);
-	//mutex_unlock(&dmxdev->mutex);
-	return ret;
+	return dvb_dmxdev_buffer_read(&dmxdev->dvr_buffer,
+				      file->f_flags & O_NONBLOCK,
+				      buf, count, ppos);
 }
 
 static int dvb_dvr_set_buffer_size(struct dmxdev *dmxdev,
_

Re: [PATCH] dvb-core: Fix potential mutex_unlock without mutex_lock in dvb_dvr_read

[Simon Arlott]

On 30/04/09 21:18, Andrew Morton wrote:
> On Thu, 23 Apr 2009 18:32:13 +0100
> Simon Arlott <simon@fire.lp0.eu> wrote:
> 
>> dvb_dvr_read may unlock the dmxdev mutex and return -ENODEV,
>> except this function is a file op and will never be called
>> with the mutex held.
>> 
>> There's existing mutex_lock and mutex_unlock around the actual
>> read but it's commented out. These should probably be uncommented
>> but the read blocks and this could block another non-blocking
>> reader on the mutex instead.
>> 
>> This change comments out the extra mutex_unlock.
>> 
>> Signed-off-by: Simon Arlott <simon@fire.lp0.eu>
>> ---
>> This has been on my TODO list for far too long... I did come
>> up with a mutex_trylock/mutex_lock_interruptible version but
>> claiming that it'll block when it may not doesn't make sense
>> (and any blocking read would cause all non-blocking reads to
>> continually return -EWOULDBLOCK until there is data).
>> 
>>  drivers/media/dvb/dvb-core/dmxdev.c |    2 +-
>>  1 files changed, 1 insertions(+), 1 deletions(-)
>> 
>> diff --git a/drivers/media/dvb/dvb-core/dmxdev.c b/drivers/media/dvb/dvb-core/dmxdev.c
>> index c35fbb8..d6d098a 100644
>> --- a/drivers/media/dvb/dvb-core/dmxdev.c
>> +++ b/drivers/media/dvb/dvb-core/dmxdev.c
>> @@ -247,7 +247,7 @@ static ssize_t dvb_dvr_read(struct file *file, char __user *buf, size_t count,
>>  	int ret;
>>  
>>  	if (dmxdev->exit) {
>> -		mutex_unlock(&dmxdev->mutex);
>> +		//mutex_unlock(&dmxdev->mutex);
>>  		return -ENODEV;
>>  	}
> 
> Is there any value in retaining all the commented-out lock operations,
> or can we zap 'em?

I'm assuming they should really be there - it's just not practical
because the call to dvb_dmxdev_buffer_read is likely to block waiting
for data.
 
> --- a/drivers/media/dvb/dvb-core/dmxdev.c~dvb-core-fix-potential-mutex_unlock-without-mutex_lock-in-dvb_dvr_read-fix
> +++ a/drivers/media/dvb/dvb-core/dmxdev.c
> @@ -244,19 +244,13 @@ static ssize_t dvb_dvr_read(struct file 
>  {
>  	struct dvb_device *dvbdev = file->private_data;
>  	struct dmxdev *dmxdev = dvbdev->priv;
> -	int ret;
>  
> -	if (dmxdev->exit) {
> -		//mutex_unlock(&dmxdev->mutex);
> +	if (dmxdev->exit)
>  		return -ENODEV;
> -	}
>  
> -	//mutex_lock(&dmxdev->mutex);
> -	ret = dvb_dmxdev_buffer_read(&dmxdev->dvr_buffer,
> -				     file->f_flags & O_NONBLOCK,
> -				     buf, count, ppos);
> -	//mutex_unlock(&dmxdev->mutex);
> -	return ret;
> +	return dvb_dmxdev_buffer_read(&dmxdev->dvr_buffer,
> +				      file->f_flags & O_NONBLOCK,
> +				      buf, count, ppos);
>  }
>  
>  static int dvb_dvr_set_buffer_size(struct dmxdev *dmxdev,
> _
> 


-- 
Simon Arlott

Re: [PATCH] dvb-core: Fix potential mutex_unlock without mutex_lock in dvb_dvr_read

[Andrew Morton]

On Thu, 30 Apr 2009 22:42:06 +0100
Simon Arlott <simon@fire.lp0.eu> wrote:

> >> diff --git a/drivers/media/dvb/dvb-core/dmxdev.c b/drivers/media/dvb/dvb-core/dmxdev.c
> >> index c35fbb8..d6d098a 100644
> >> --- a/drivers/media/dvb/dvb-core/dmxdev.c
> >> +++ b/drivers/media/dvb/dvb-core/dmxdev.c
> >> @@ -247,7 +247,7 @@ static ssize_t dvb_dvr_read(struct file *file, char __user *buf, size_t count,
> >>  	int ret;
> >>  
> >>  	if (dmxdev->exit) {
> >> -		mutex_unlock(&dmxdev->mutex);
> >> +		//mutex_unlock(&dmxdev->mutex);
> >>  		return -ENODEV;
> >>  	}
> > 
> > Is there any value in retaining all the commented-out lock operations,
> > or can we zap 'em?
> 
> I'm assuming they should really be there - it's just not practical
> because the call to dvb_dmxdev_buffer_read is likely to block waiting
> for data.

well..  such infomation is much better communicated via a nice comment,
rather than mystery-dead-code?

Re: [PATCH] dvb-core: Fix potential mutex_unlock without mutex_lock in dvb_dvr_read

[Devin Heitmueller]

On Thu, Apr 30, 2009 at 5:48 PM, Andrew Morton
<akpm@linux-foundation.org> wrote:
> On Thu, 30 Apr 2009 22:42:06 +0100
> Simon Arlott <simon@fire.lp0.eu> wrote:
>
>> >> diff --git a/drivers/media/dvb/dvb-core/dmxdev.c b/drivers/media/dvb/dvb-core/dmxdev.c
>> >> index c35fbb8..d6d098a 100644
>> >> --- a/drivers/media/dvb/dvb-core/dmxdev.c
>> >> +++ b/drivers/media/dvb/dvb-core/dmxdev.c
>> >> @@ -247,7 +247,7 @@ static ssize_t dvb_dvr_read(struct file *file, char __user *buf, size_t count,
>> >>    int ret;
>> >>
>> >>    if (dmxdev->exit) {
>> >> -          mutex_unlock(&dmxdev->mutex);
>> >> +          //mutex_unlock(&dmxdev->mutex);
>> >>            return -ENODEV;
>> >>    }
>> >
>> > Is there any value in retaining all the commented-out lock operations,
>> > or can we zap 'em?
>>
>> I'm assuming they should really be there - it's just not practical
>> because the call to dvb_dmxdev_buffer_read is likely to block waiting
>> for data.
>
> well..  such infomation is much better communicated via a nice comment,
> rather than mystery-dead-code?

I'm doing some review of the locking in dvb core as a result of a race
condition I found earlier in the week.  I'll take a look at this too
when I get a few minutes.

Devin

-- 
Devin J. Heitmueller
http://www.devinheitmueller.com
AIM: devinheitmueller