Message-ID: <CAJedcCyVOpNyXsm6NW738tPvGF_nNDh_NCKXqKWgrnhKV5y-2g@mail.gmail.com>
Date:   Wed, 8 Mar 2023 17:11:21 +0800
From:   Zheng Hacker <hackerzheng666@...il.com>
To:     Kyrie Wu (吴晗) <Kyrie.Wu@...iatek.com>
Cc:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-mediatek@...ts.infradead.org" 
        <linux-mediatek@...ts.infradead.org>,
        "linux-media@...r.kernel.org" <linux-media@...r.kernel.org>,
        "mchehab@...nel.org" <mchehab@...nel.org>,
        "zyytlz.wz@....com" <zyytlz.wz@....com>,
        Bin Liu (刘彬) <bin.liu@...iatek.com>,
        "alex000young@...il.com" <alex000young@...il.com>,
        Irui Wang (王瑞) <Irui.Wang@...iatek.com>,
        "linux-arm-kernel@...ts.infradead.org" 
        <linux-arm-kernel@...ts.infradead.org>,
        "matthias.bgg@...il.com" <matthias.bgg@...il.com>,
        "angelogioacchino.delregno@...labora.com" 
        <angelogioacchino.delregno@...labora.com>,
        "1395428693sheep@...il.com" <1395428693sheep@...il.com>
Subject: Re: [RESEND PATCH] media: mtk-jpeg: Fix use after free bug due to
 uncanceled work

Dear Kyrie,

Sorry for my misunderstanding. I've seen the timeout worker is bound
with mtk_jpegdec_timeout_work rather than mtk_jpeg_job_timeout_work.So
the competition won't happen.

Back to the beginning scene, I still don't know if it's a UAF issue or
not. The normal schedule is very strong to me. If we can call
mtk_jpeg_remove directly without calling mtk_jpeg_release,
The UAF is still possible. Otherwis, I think it's safe here.

Best regards,
Zheng

Kyrie Wu (吴晗) <Kyrie.Wu@...iatek.com> 于2023年3月8日周三 15:39写道：
>
> On Wed, 2023-03-08 at 14:10 +0800, Zheng Hacker wrote:
> > Hi Kyrie,
> >
> > After reviewing the code, I found anothe possible code path. As I am
> > not familiar with the module. It has high possibility it's wrong.
> > Could please help me check this? Very much appreciated for your
> > valuable time.
> >
> > In summary, mtk_jpegdec_worker was set in mtk_jpeg_open and started
> > in
> > mtk_jpeg_multicore_dec_device_run, which made it running on cpu1.
> > Inside the mtk_jpeg_multicore_dec_device_run, it will call
> > schedule_delayed_work  to start the timeout_work, which will make it
> > running on cpu2. Meanwhile, we can call
> > mtk_jpeg_release to cancel the job. But there might be a race between
> > mtk_jpegdec_worker and v4l2_m2m_cancel_job. It may call
> > v4l2_m2m_job_finish too early to wake up the event.
> > The remove will go on, the other race is as described earlier.
> >
> > cpu0                         cpu1                  cpu2
> > (1)->device_run
> > mtk_jpeg_multicore
> > _dec_device_run
> > queue_work
> > (jpeg->workqueue,
> > &ctx->jpeg_work);
> >                       (2)mtk_jpegdec_worker
> > (3)mtk_jpeg_release
> > v4l2_m2m_cancel_job
> > wait event
> >
> >                       schedule_delayed_work
> >                                             (4)mtk_jpeg_job_timeout_w
> > ork
> >                       (5)v4l2_m2m_job_finish
> >                       wake up
> > (6)mtk_jpeg_remove
> > v4l2_m2m_release
> > kfree(m2m_dev)
> >                                             (7)v4l2_m2m_get_curr_priv
> >
> Dear zheng,
>
> The mtk_jpeg_multicore_dec_device_run function is used for multi-hw
> jpeg decoding. Instead of scheduling mtk_jpeg_job_timeout_work,
> mtk_jpegdec_worker is scheduled in this function.
>
> The mtk_jpeg_dec_device_run function is used for single hw jpeg
> decoding, which schedules mtk_jpeg_job_timeout_work.
>
> A driver is either a single hw driver or a multi-hw driver and cannot
> represent both at the same time.
> mtk_jpeg_job_timeout_work and mtk_jpegdec_worker cannot be scheduled at
> the same time.
> So mtk_jpeg_job_timeout_work calls v4l2_m2m_job_finish would not cause
> competition between the mtk_jpegdec_worker and v4l2_m2m_cancel_job.
>
> Regards,
> Kyrie.
>
>
> > Kyrie Wu (吴晗) <Kyrie.Wu@...iatek.com> 于2023年3月8日周三 11:32写道：
> > >
> > > On Wed, 2023-03-08 at 10:20 +0800, Zheng Hacker wrote:
> > > > Hi Kyrie,
> > > >
> > > > Thank you for your careful analysis and response. I still have
> > > > some
> > > > areas that I don't quite understand and would like to ask for
> > > > clarification. That is, how do the function pointers for stop
> > > > streaming, initialized as mtk_jpeg_enc_stop_streaming and
> > > > mtk_jpeg_dec_stop_streaming, ensure that the worker is canceled?
> > > > I
> > > > would greatly appreciate your response.
> > > >
> > > > Best regards,
> > > > Zheng
> > >
> > > Dear zheng,
> > >
> > > For stop streaming, what I mean is that stoppping jpeg decoding or
> > > encoding.
> > > Ok, let me introduce the sw flow of stop streaming:
> > > Firstly, the app will call v4l2_m2m_ioctl_streamoff, which will
> > > call
> > > v4l2_m2m_cancel_job, if it finds a job running(as you note, cpu1 is
> > > running), it will wait event, the event is wake up by
> > > v4l2_m2m_job_finish function. And v4l2_m2m_job_finish is called by
> > > jpeg
> > > dec/enc irq handler, which means that the waitting would result mtk
> > > hw
> > > to finish dec/enc, irq will occur and irq handler would cancel
> > > timeout
> > > worker. The follow is shown as blow.
> > > v4l2_m2m_ioctl_streamoff
> > >    v4l2_m2m_cancel_job                mtk_jpeg_enc_irq/mtk_jpeg_dec
> > > _irq
> > >        wait evnet <------ wake up ------v4l2_m2m_job_finish
> > >                                         cancel timeout work
> > >
> > > As mentioned above, if it is normal stop streaming action, there
> > > will
> > > be no happen that the timeout worker does not canceled.
> > >
> > > But if mtk_jpeg_remove is called directly without above flow, it
> > > would
> > > cause lots of issues.
> > >
> > > Regards,
> > > Kyrie.
> > > >
> > > > Kyrie Wu (吴晗) <Kyrie.Wu@...iatek.com> 于2023年3月8日周三 10:02写道：
> > > > >
> > > > > On Tue, 2023-03-07 at 23:03 +0800, Zheng Hacker wrote:
> > > > > > The timer function was set in mtk_jpeg_probe with
> > > > > > mtk_jpeg_job_timeout_work function.
> > > > > > And the worker is started in mtk_jpeg_dec_device_run.
> > > > > > There are two functions (mtk_jpeg_enc_irq and
> > > > > > mtk_jpeg_dec_irq)
> > > > > > which
> > > > > > may cancel the worker.
> > > > > > They are used as IRQ handler function which is saved as
> > > > > > function
> > > > > > pointer in a variable.
> > > > > > In mtk_jpeg_probe, they are registered by devm_request_irq:
> > > > > >
> > > > > > ret = devm_request_irq(&pdev->dev,
> > > > > >                jpeg_irq,
> > > > > >                jpeg->variant->irq_handler,
> > > > > >                0,
> > > > > >                pdev->name, jpeg);
> > > > > >     if (ret) {
> > > > > >       dev_err(&pdev->dev, "Failed to request jpeg_irq %d
> > > > > > (%d)\n",
> > > > > >         jpeg_irq, ret);
> > > > > >       return ret;
> > > > > >     }
> > > > > >
> > > > > > However, if we remove the module without triggering the irq,
> > > > > > the
> > > > > > worker will never be removed.
> > > > > >
> > > > > > As for the schedule, mtk_jpeg_dec_device_run and
> > > > > > mtk_jpeg_enc_device_run will start the worker.
> > > > > > The schedule invoking is quite complicated. As far as I know,
> > > > > > the
> > > > > > invoking chain is as follows:
> > > > > >
> > > > > > v4l2_m2m_init->v4l2_m2m_device_run_work->v4l2_m2m_try_run
> > > > > >
> > > > > > the v4l2_m2m_device_run_work function is also a worker which
> > > > > > is
> > > > > > set
> > > > > > in
> > > > > > v4l2_m2m_init and started in
> > > > > > v4l2_m2m_schedule_next_job.
> > > > > >
> > > > > > Before calling remove function, the  mtk_jpeg_release was
> > > > > > invoked
> > > > > > to
> > > > > > release the related resource.
> > > > > >
> > > > > > v4l2_m2m_cancel_job will cancel the job by calling
> > > > > > m2m_dev->m2m_ops->job_abort(m2m_ctx->priv).
> > > > > >
> > > > > > But this will only cancel the current queue by
> > > > > > list_del(&m2m_dev->curr_ctx->queue);
> > > > > >
> > > > > > I think this can not cancel the posted task mentioned before.
> > > > > > So
> > > > > > I
> > > > > > think if mtk_jpeg_job_timeout_work
> > > > > >
> > > > > > is working on, and using jpeg->m2m_dev after freeing it in
> > > > > > mtk_jpeg_remove, it will cause a UAF bug.
> > > > > >
> > > > > > static int mtk_jpeg_release(struct file *file)
> > > > > > {
> > > > > >   struct mtk_jpeg_dev *jpeg = video_drvdata(file);
> > > > > >   struct mtk_jpeg_ctx *ctx = mtk_jpeg_fh_to_ctx(file-
> > > > > > > private_data);
> > > > > >
> > > > > >   mutex_lock(&jpeg->lock);
> > > > > >   v4l2_ctrl_handler_free(&ctx->ctrl_hdl);
> > > > > >   [1] v4l2_m2m_ctx_release(ctx->fh.m2m_ctx);
> > > > > >   v4l2_fh_del(&ctx->fh);
> > > > > >   v4l2_fh_exit(&ctx->fh);
> > > > > >   kfree(ctx);
> > > > > >   mutex_unlock(&jpeg->lock);
> > > > > >   return 0;
> > > > > > }
> > > > > >
> > > > > > void v4l2_m2m_ctx_release(struct v4l2_m2m_ctx *m2m_ctx)
> > > > > > {
> > > > > >   /* wait until the current context is dequeued from
> > > > > > job_queue */
> > > > > >   [2] v4l2_m2m_cancel_job(m2m_ctx);
> > > > > >
> > > > > >   vb2_queue_release(&m2m_ctx->cap_q_ctx.q);
> > > > > >   vb2_queue_release(&m2m_ctx->out_q_ctx.q);
> > > > > >
> > > > > >   kfree(m2m_ctx);
> > > > > > }
> > > > > >
> > > > > > Note that all of this is static analysis, which may be false
> > > > > > positive.
> > > > > > Feel free to tell me if there is something I've missed.
> > > > > >
> > > > > > Regard,
> > > > > > Zheng
> > > > >
> > > > > Dear Zheng,
> > > > >
> > > > > You set up an application scenario:
> > > > > cpu1 is using the mtk-jpeg driver and timeout work has been
> > > > > scheduled.
> > > > > At the same time cpu0 wanted to remove the mtk-jpeg driver,
> > > > > which
> > > > > caused the UAF bug.
> > > > > I wonder if such an irrational application scenario could
> > > > > exist.
> > > > > This
> > > > > scenario, as you described, not only leads to the problems you
> > > > > mentioned, but also to output&capture memory leaks and
> > > > > unreleased
> > > > > resources, such as spinlock.
> > > > > Typically, if we want to remove the driver, we firstly do stop
> > > > > streaming, which ensures that the worker has been canceled.
> > > > > I agree with your changes from the perspective of strengthening
> > > > > the
> > > > > robustness of the driver code.
> > > > >
> > > > > Regards,
> > > > > Kyrie.
> > > > > >
> > > > > > Irui Wang (王瑞) <Irui.Wang@...iatek.com> 于2023年3月7日周二 18:23写道：
> > > > > > >
> > > > > > > Dear Angelo and Zheng,
> > > > > > >
> > > > > > > Thanks for your patch and comments.
> > > > > > >
> > > > > > > Dear Kyrie,
> > > > > > >
> > > > > > > Please help to check this, thanks.
> > > > > > >
> > > > > > > Best Regards
> > > > > > >
> > > > > > > On Tue, 2023-03-07 at 10:49 +0100, AngeloGioacchino Del
> > > > > > > Regno
> > > > > > > wrote:
> > > > > > > > Il 07/03/23 10:27, Zheng Hacker ha scritto:
> > > > > > > > > Hi,
> > > > > > > > >
> > > > > > > > > Is there anyone who can help with this? I can provide
> > > > > > > > > more
> > > > > > > > > details
> > > > > > > > > like invoking chain if needed.
> > > > > > > > >
> > > > > > > >
> > > > > > > > Providing more details is always good. Please do.
> > > > > > > >
> > > > > > > > Meanwhile, adding Irui Wang to the loop: he's doing mtk-
> > > > > > > > jpeg.
> > > > > > > >
> > > > > > > > Regards,
> > > > > > > > Angelo
> > > > > > > >
> > > > > > > > > Thanks,
> > > > > > > > > Zheng
> > > > > > > > >
> > > > > > > > > Zheng Wang <zyytlz.wz@....com> 于2023年3月6日周一 14:28写道：
> > > > > > > > > >
> > > > > > > > > > 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 <zyytlz.wz@....com>
> > > > > > > > > > ---
> > > > > > > > > >
> > > > > > > > > > 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
> > > > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >

Powered by blists - more mailing lists