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 <[email protected]>
---
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
On Thu, 23 Apr 2009 18:32:13 +0100
Simon Arlott <[email protected]> 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 <[email protected]>
> ---
> 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,
_
On 30/04/09 21:18, Andrew Morton wrote:
> On Thu, 23 Apr 2009 18:32:13 +0100
> Simon Arlott <[email protected]> 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 <[email protected]>
>> ---
>> 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
On Thu, 30 Apr 2009 22:42:06 +0100
Simon Arlott <[email protected]> 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?
On Thu, Apr 30, 2009 at 5:48 PM, Andrew Morton
<[email protected]> wrote:
> On Thu, 30 Apr 2009 22:42:06 +0100
> Simon Arlott <[email protected]> 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