In mtk_jpeg_probe, &jpeg->job_timeout_work is bound with
mtk_jpeg_job_timeout_work. Then mtk_jpeg_dec_device_run
and mtk_jpeg_enc_device_run may be called to start the
work.
If we remove the module which will call mtk_jpeg_remove
to make cleanup, there may be a unfinished work. The
possible sequence is as follows, which will cause a
typical UAF bug.
Fix it by canceling the work before cleanup in the mtk_jpeg_remove
CPU0 CPU1
|mtk_jpeg_job_timeout_work
mtk_jpeg_remove |
v4l2_m2m_release |
kfree(m2m_dev); |
|
| v4l2_m2m_get_curr_priv
| m2m_dev->curr_ctx //use
Signed-off-by: Zheng Wang <[email protected]>
---
drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
index 969516a940ba..364513e7897e 100644
--- a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
+++ b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
@@ -1793,7 +1793,7 @@ static int mtk_jpeg_probe(struct platform_device *pdev)
static int mtk_jpeg_remove(struct platform_device *pdev)
{
struct mtk_jpeg_dev *jpeg = platform_get_drvdata(pdev);
-
+ cancel_delayed_work(&jpeg->job_timeout_work);
pm_runtime_disable(&pdev->dev);
video_unregister_device(jpeg->vdev);
v4l2_m2m_release(jpeg->m2m_dev);
--
2.25.1
On Thu, Mar 02, 2023 at 05:37:15PM +0800, Zheng Wang wrote:
> In mtk_jpeg_probe, &jpeg->job_timeout_work is bound with
> mtk_jpeg_job_timeout_work. Then mtk_jpeg_dec_device_run
> and mtk_jpeg_enc_device_run may be called to start the
> work.
> If we remove the module which will call mtk_jpeg_remove
> to make cleanup, there may be a unfinished work. The
> possible sequence is as follows, which will cause a
> typical UAF bug.
>
> Fix it by canceling the work before cleanup in the mtk_jpeg_remove
>
> CPU0 CPU1
>
> |mtk_jpeg_job_timeout_work
> mtk_jpeg_remove |
> v4l2_m2m_release |
> kfree(m2m_dev); |
> |
> | v4l2_m2m_get_curr_priv
> | m2m_dev->curr_ctx //use
>
> Signed-off-by: Zheng Wang <[email protected]>
> ---
> drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
> index 969516a940ba..364513e7897e 100644
> --- a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
> +++ b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
> @@ -1793,7 +1793,7 @@ static int mtk_jpeg_probe(struct platform_device *pdev)
> static int mtk_jpeg_remove(struct platform_device *pdev)
> {
> struct mtk_jpeg_dev *jpeg = platform_get_drvdata(pdev);
> -
> + cancel_delayed_work(&jpeg->job_timeout_work);
The empty line is needed (coding style). Also, this doesn't cancel
the worker if it is already running. This should probably be
cancel_delayed_work_sync(). Even then the question is if it is
possible that new work is queued before the device is unregistered.
Guenter
> pm_runtime_disable(&pdev->dev);
> video_unregister_device(jpeg->vdev);
> v4l2_m2m_release(jpeg->m2m_dev);
> --
> 2.25.1
>
Hi,
Thanks for your reply. I think you're right. I don't know if there is
other method to stop new work from enqueing. Could you please give me
some advice about the fix?
Regards,
Zheng
Guenter Roeck <[email protected]> 于2023年3月9日周四 08:27写道：
>
> On Thu, Mar 02, 2023 at 05:37:15PM +0800, Zheng Wang wrote:
> > In mtk_jpeg_probe, &jpeg->job_timeout_work is bound with
> > mtk_jpeg_job_timeout_work. Then mtk_jpeg_dec_device_run
> > and mtk_jpeg_enc_device_run may be called to start the
> > work.
> > If we remove the module which will call mtk_jpeg_remove
> > to make cleanup, there may be a unfinished work. The
> > possible sequence is as follows, which will cause a
> > typical UAF bug.
> >
> > Fix it by canceling the work before cleanup in the mtk_jpeg_remove
> >
> > CPU0 CPU1
> >
> > |mtk_jpeg_job_timeout_work
> > mtk_jpeg_remove |
> > v4l2_m2m_release |
> > kfree(m2m_dev); |
> > |
> > | v4l2_m2m_get_curr_priv
> > | m2m_dev->curr_ctx //use
> >
> > Signed-off-by: Zheng Wang <[email protected]>
> > ---
> > drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
> > index 969516a940ba..364513e7897e 100644
> > --- a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
> > +++ b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
> > @@ -1793,7 +1793,7 @@ static int mtk_jpeg_probe(struct platform_device *pdev)
> > static int mtk_jpeg_remove(struct platform_device *pdev)
> > {
> > struct mtk_jpeg_dev *jpeg = platform_get_drvdata(pdev);
> > -
> > + cancel_delayed_work(&jpeg->job_timeout_work);
>
> The empty line is needed (coding style). Also, this doesn't cancel
> the worker if it is already running. This should probably be
> cancel_delayed_work_sync(). Even then the question is if it is
> possible that new work is queued before the device is unregistered.
>
> Guenter
>
> > pm_runtime_disable(&pdev->dev);
> > video_unregister_device(jpeg->vdev);
> > v4l2_m2m_release(jpeg->m2m_dev);
> > --
> > 2.25.1
> >
On 3/8/23 19:58, Zheng Hacker wrote:
> Hi,
>
> Thanks for your reply. I think you're right. I don't know if there is
> other method to stop new work from enqueing. Could you please give me
> some advice about the fix?
>
Top-posting is discouraged.
Anyway -
I don't know the code well enough to suggest a solution.
It all depends on the driver architecture. The maintainers might
have a better idea.
A worse problem appears to be that the worker is also canceled
from mtk_jpeg_enc_irq() and mtk_jpeg_dec_irq(). Those are non-threaded
interrupt handlers which, as far as I know, must not sleep and thus
can not call cancel_delayed_work_sync(). I have no idea how to solve
that problem either.
Guenter
> Regards,
> Zheng
>
> Guenter Roeck <[email protected]> 于2023年3月9日周四 08:27写道：
>>
>> On Thu, Mar 02, 2023 at 05:37:15PM +0800, Zheng Wang wrote:
>>> In mtk_jpeg_probe, &jpeg->job_timeout_work is bound with
>>> mtk_jpeg_job_timeout_work. Then mtk_jpeg_dec_device_run
>>> and mtk_jpeg_enc_device_run may be called to start the
>>> work.
>>> If we remove the module which will call mtk_jpeg_remove
>>> to make cleanup, there may be a unfinished work. The
>>> possible sequence is as follows, which will cause a
>>> typical UAF bug.
>>>
>>> Fix it by canceling the work before cleanup in the mtk_jpeg_remove
>>>
>>> CPU0 CPU1
>>>
>>> |mtk_jpeg_job_timeout_work
>>> mtk_jpeg_remove |
>>> v4l2_m2m_release |
>>> kfree(m2m_dev); |
>>> |
>>> | v4l2_m2m_get_curr_priv
>>> | m2m_dev->curr_ctx //use
>>>
>>> Signed-off-by: Zheng Wang <[email protected]>
>>> ---
>>> drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
>>> index 969516a940ba..364513e7897e 100644
>>> --- a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
>>> +++ b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
>>> @@ -1793,7 +1793,7 @@ static int mtk_jpeg_probe(struct platform_device *pdev)
>>> static int mtk_jpeg_remove(struct platform_device *pdev)
>>> {
>>> struct mtk_jpeg_dev *jpeg = platform_get_drvdata(pdev);
>>> -
>>> + cancel_delayed_work(&jpeg->job_timeout_work);
>>
>> The empty line is needed (coding style). Also, this doesn't cancel
>> the worker if it is already running. This should probably be
>> cancel_delayed_work_sync(). Even then the question is if it is
>> possible that new work is queued before the device is unregistered.
>>
>> Guenter
>>
>>> pm_runtime_disable(&pdev->dev);
>>> video_unregister_device(jpeg->vdev);
>>> v4l2_m2m_release(jpeg->m2m_dev);
>>> --
>>> 2.25.1
>>>
Guenter Roeck <[email protected]> 于2023年3月9日周四 13:30写道：
>
> On 3/8/23 19:58, Zheng Hacker wrote:
> > Hi,
> >
> > Thanks for your reply. I think you're right. I don't know if there is
> > other method to stop new work from enqueing. Could you please give me
> > some advice about the fix?
> >
>
> Top-posting is discouraged.
>
Sorry I forgot that. Thanks for the kind reminder.
> Anyway -
> I don't know the code well enough to suggest a solution.
> It all depends on the driver architecture. The maintainers might
> have a better idea.
>
Yes, some related developers have reached out to me and discussed fixes with me.
> A worse problem appears to be that the worker is also canceled
> from mtk_jpeg_enc_irq() and mtk_jpeg_dec_irq(). Those are non-threaded
> interrupt handlers which, as far as I know, must not sleep and thus
> can not call cancel_delayed_work_sync(). I have no idea how to solve
> that problem either.
>
I'd be glad to pass along your thoughts and recommendations to the
relevant parties.
Best regards,
Zheng