* [PATCH] media: fix media_device_unregister() to destroy media device device resource
@ 2016-03-17 22:46 Shuah Khan
2016-03-18 9:52 ` Mauro Carvalho Chehab
0 siblings, 1 reply; 3+ messages in thread
From: Shuah Khan @ 2016-03-17 22:46 UTC (permalink / raw)
To: mchehab; +Cc: Shuah Khan, tiwai, linux-media, linux-kernel
When all drivers except usb-core driver is unbound, destroy the media device
resource. Other wise, media device resource will persist in a defunct state.
This leads to use-after-free and bad access errors during a subsequent bind.
Fix it to destroy the media device resource when last reference is released
in media_device_unregister().
Signed-off-by: Shuah Khan <shuahkh@osg.samsung.com>
---
drivers/media/media-device.c | 28 ++++++++++++++++++++++------
1 file changed, 22 insertions(+), 6 deletions(-)
diff --git a/drivers/media/media-device.c b/drivers/media/media-device.c
index 070421e..7312612 100644
--- a/drivers/media/media-device.c
+++ b/drivers/media/media-device.c
@@ -822,22 +822,38 @@ printk("%s: mdev=%p\n", __func__, mdev);
dev_dbg(mdev->dev, "Media device unregistered\n");
}
+static void media_device_release_devres(struct device *dev, void *res)
+{
+}
+
+static void media_device_destroy_devres(struct device *dev)
+{
+ int ret;
+
+ ret = devres_destroy(dev, media_device_release_devres, NULL, NULL);
+ pr_debug("%s: devres_destroy() returned %d\n", __func__, ret);
+}
+
void media_device_unregister(struct media_device *mdev)
{
+ int ret;
+ struct device *dev;
printk("%s: mdev=%p\n", __func__, mdev);
if (mdev == NULL)
return;
- mutex_lock(&mdev->graph_mutex);
- kref_put(&mdev->kref, do_media_device_unregister);
- mutex_unlock(&mdev->graph_mutex);
+ ret = kref_put_mutex(&mdev->kref, do_media_device_unregister,
+ &mdev->graph_mutex);
+ if (ret) {
+ /* do_media_device_unregister() has run */
+ dev = mdev->dev;
+ mutex_unlock(&mdev->graph_mutex);
+ media_device_destroy_devres(dev);
+ }
}
EXPORT_SYMBOL_GPL(media_device_unregister);
-static void media_device_release_devres(struct device *dev, void *res)
-{
-}
struct media_device *media_device_get_devres(struct device *dev)
{
--
2.5.0
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH] media: fix media_device_unregister() to destroy media device device resource
2016-03-17 22:46 [PATCH] media: fix media_device_unregister() to destroy media device device resource Shuah Khan
@ 2016-03-18 9:52 ` Mauro Carvalho Chehab
2016-03-18 13:36 ` Shuah Khan
0 siblings, 1 reply; 3+ messages in thread
From: Mauro Carvalho Chehab @ 2016-03-18 9:52 UTC (permalink / raw)
To: Shuah Khan; +Cc: tiwai, linux-media, linux-kernel
Em Thu, 17 Mar 2016 16:46:36 -0600
Shuah Khan <shuahkh@osg.samsung.com> escreveu:
> When all drivers except usb-core driver is unbound, destroy the media device
> resource. Other wise, media device resource will persist in a defunct state.
> This leads to use-after-free and bad access errors during a subsequent bind.
> Fix it to destroy the media device resource when last reference is released
> in media_device_unregister().
>
> Signed-off-by: Shuah Khan <shuahkh@osg.samsung.com>
> ---
> drivers/media/media-device.c | 28 ++++++++++++++++++++++------
> 1 file changed, 22 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/media/media-device.c b/drivers/media/media-device.c
> index 070421e..7312612 100644
> --- a/drivers/media/media-device.c
> +++ b/drivers/media/media-device.c
> @@ -822,22 +822,38 @@ printk("%s: mdev=%p\n", __func__, mdev);
> dev_dbg(mdev->dev, "Media device unregistered\n");
> }
>
> +static void media_device_release_devres(struct device *dev, void *res)
> +{
> +}
> +
> +static void media_device_destroy_devres(struct device *dev)
> +{
> + int ret;
> +
> + ret = devres_destroy(dev, media_device_release_devres, NULL, NULL);
> + pr_debug("%s: devres_destroy() returned %d\n", __func__, ret);
> +}
> +
> void media_device_unregister(struct media_device *mdev)
> {
> + int ret;
> + struct device *dev;
> printk("%s: mdev=%p\n", __func__, mdev);
> if (mdev == NULL)
> return;
>
> - mutex_lock(&mdev->graph_mutex);
> - kref_put(&mdev->kref, do_media_device_unregister);
> - mutex_unlock(&mdev->graph_mutex);
> + ret = kref_put_mutex(&mdev->kref, do_media_device_unregister,
> + &mdev->graph_mutex);
> + if (ret) {
> + /* do_media_device_unregister() has run */
> + dev = mdev->dev;
> + mutex_unlock(&mdev->graph_mutex);
> + media_device_destroy_devres(dev);
This doesn't seem right: what happens on drivers that don't use
devres to allocate struct media_device?
> + }
>
> }
> EXPORT_SYMBOL_GPL(media_device_unregister);
>
> -static void media_device_release_devres(struct device *dev, void *res)
> -{
> -}
>
> struct media_device *media_device_get_devres(struct device *dev)
> {
--
Thanks,
Mauro
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] media: fix media_device_unregister() to destroy media device device resource
2016-03-18 9:52 ` Mauro Carvalho Chehab
@ 2016-03-18 13:36 ` Shuah Khan
0 siblings, 0 replies; 3+ messages in thread
From: Shuah Khan @ 2016-03-18 13:36 UTC (permalink / raw)
To: Mauro Carvalho Chehab; +Cc: tiwai, linux-media, linux-kernel, Shuah Khan
On 03/18/2016 03:52 AM, Mauro Carvalho Chehab wrote:
> Em Thu, 17 Mar 2016 16:46:36 -0600
> Shuah Khan <shuahkh@osg.samsung.com> escreveu:
>
>> When all drivers except usb-core driver is unbound, destroy the media device
>> resource. Other wise, media device resource will persist in a defunct state.
>> This leads to use-after-free and bad access errors during a subsequent bind.
>> Fix it to destroy the media device resource when last reference is released
>> in media_device_unregister().
>>
>> Signed-off-by: Shuah Khan <shuahkh@osg.samsung.com>
>> ---
>> drivers/media/media-device.c | 28 ++++++++++++++++++++++------
>> 1 file changed, 22 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/media/media-device.c b/drivers/media/media-device.c
>> index 070421e..7312612 100644
>> --- a/drivers/media/media-device.c
>> +++ b/drivers/media/media-device.c
>> @@ -822,22 +822,38 @@ printk("%s: mdev=%p\n", __func__, mdev);
>> dev_dbg(mdev->dev, "Media device unregistered\n");
>> }
>>
>> +static void media_device_release_devres(struct device *dev, void *res)
>> +{
>> +}
>> +
>> +static void media_device_destroy_devres(struct device *dev)
>> +{
>> + int ret;
>> +
>> + ret = devres_destroy(dev, media_device_release_devres, NULL, NULL);
>> + pr_debug("%s: devres_destroy() returned %d\n", __func__, ret);
>> +}
>> +
>> void media_device_unregister(struct media_device *mdev)
>> {
>> + int ret;
>> + struct device *dev;
>> printk("%s: mdev=%p\n", __func__, mdev);
>> if (mdev == NULL)
>> return;
>>
>> - mutex_lock(&mdev->graph_mutex);
>> - kref_put(&mdev->kref, do_media_device_unregister);
>> - mutex_unlock(&mdev->graph_mutex);
>> + ret = kref_put_mutex(&mdev->kref, do_media_device_unregister,
>> + &mdev->graph_mutex);
>> + if (ret) {
>> + /* do_media_device_unregister() has run */
>> + dev = mdev->dev;
>> + mutex_unlock(&mdev->graph_mutex);
>
>
>> + media_device_destroy_devres(dev);
>
> This doesn't seem right: what happens on drivers that don't use
> devres to allocate struct media_device?
>
That is okay. devres_destroy() won't find the resource. The way it works
is it will try to find the resource with the match routine and data and
that step will fail it will return -ENOENT. At that point nothing more
is done.
ret = devres_destroy(dev, media_device_release_devres, NULL, NULL);
pr_debug("%s: devres_destroy() returned %d\n", __func__, ret);
devres_destroy() combines the devres_find() and remove. So we are good
here.
thanks,
-- Shuah
--
Shuah Khan
Sr. Linux Kernel Developer
Open Source Innovation Group
Samsung Research America (Silicon Valley)
shuahkh@osg.samsung.com | (970) 217-8978
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2016-03-18 13:37 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-03-17 22:46 [PATCH] media: fix media_device_unregister() to destroy media device device resource Shuah Khan
2016-03-18 9:52 ` Mauro Carvalho Chehab
2016-03-18 13:36 ` Shuah Khan
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).