Re: [RESEND PATCH v2] media: mtk-jpeg: Fix use after free bug due to uncanceled work

From: Dmitry Osipenko
Date: Thu Oct 19 2023 - 15:56:15 EST


On 10/8/23 12:13, Zheng Hacker wrote:
> Dmitry Osipenko <dmitry.osipenko@xxxxxxxxxxxxx> 于2023年9月20日周三 02:24写道:
>>
>> On 8/31/23 11:18, Zheng Hacker wrote:
>>>> The v4l2_m2m_ctx_release() already should wait for the job_timeout_work
>>>> completion or for the interrupt fire. Apparently it doesn't work in
>>>> yours case. You'll need to debug why v4l job or job_timeout_work is
>>>> running after v4l2_m2m_ctx_release(), it shouldn't happen.
>>>>
>>> Yes, v4l2_m2m_cancel_job waits for m2m_ctx->job_flags to be ~TRANS_RUNNING,
>>> the mtk_jpeg_job_timeout_work will finally invoke v4l2_m2m_job_finish
>>> to trigger that.
>>>
>>> However, this is not the only path to call v4l2_m2m_job_finish. Here
>>> is a invoking chain:
>>> v4l_streamon
>>> ->v4l2_m2m_ioctl_streamon
>>> ->v4l2_m2m_streamon
>>> ->v4l2_m2m_try_schedule
>>> ->v4l2_m2m_try_run
>>> ->mtk_jpeg_dec_device_run
>>> ->schedule_delayed_work(&jpeg->job_timeout_work...
>>> ->error path goto dec_end
>>> ->v4l2_m2m_job_finish
>>>
>>> In some specific situation, it starts the worker and also calls
>>> v4l2_m2m_job_finish, which might
>>> make v4l2_m2m_cancel_job continues.
>>
>> Then the error path should cancel the job_timeout_work, or better job
>> needs to be run after the dec/enc has been started and not before.
>>
>
> Hi,
>
> Sorry for my late reply for I just went on a long vacation.
>
> Get it. I'll write another patch and change the summary to the lack of
> canceling job in error path.
>
>> Looking further at the code, I'm confused by this hunk:
>>
>> mtk_jpeg_dec_start(comp_jpeg[hw_id]->reg_base);
>> v4l2_m2m_job_finish(jpeg->m2m_dev, ctx->fh.m2m_ctx);
>>
>> The job should be marked as finished when h/w has finished processing
>> the job and not right after the job has been started. So the job is
>> always completed and mtk_jpeg_job_timeout_work() doesn't work as
>> expected, am I missing something?
>
> After reading the code I still don't know. I didn't see any function
> like mtk_jpeg_dec_end. The same thing
> happens on mtk_jpeg_enc_start. I think I'd better fix the first
> problem and wait for someone familiar with
> the second part.

I missed that the code mentioned above is related to the multi-core hw version, while you care about single-core. Nevertheless, the multi-core device_run() looks incorrect,

So, the error code paths need to be corrected. Please try to revert yours fix and test this change:

diff --git a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
index 0051f372a66c..fd3b0587fcad 100644
--- a/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
+++ b/drivers/media/platform/mediatek/jpeg/mtk_jpeg_core.c
@@ -1254,9 +1254,6 @@ static void mtk_jpegdec_worker(struct work_struct *work)
v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);

- schedule_delayed_work(&comp_jpeg[hw_id]->job_timeout_work,
- msecs_to_jiffies(MTK_JPEG_HW_TIMEOUT_MSEC));
-
mtk_jpeg_set_dec_src(ctx, &src_buf->vb2_buf, &bs);
if (mtk_jpeg_set_dec_dst(ctx,
&jpeg_src_buf->dec_param,
@@ -1266,6 +1263,9 @@ static void mtk_jpegdec_worker(struct work_struct *work)
goto setdst_end;
}

+ schedule_delayed_work(&comp_jpeg[hw_id]->job_timeout_work,
+ msecs_to_jiffies(MTK_JPEG_HW_TIMEOUT_MSEC));
+
spin_lock_irqsave(&comp_jpeg[hw_id]->hw_lock, flags);
ctx->total_frame_num++;
mtk_jpeg_dec_reset(comp_jpeg[hw_id]->reg_base);
@@ -1330,13 +1330,13 @@ static void mtk_jpeg_dec_device_run(void *priv)
if (ret < 0)
goto dec_end;

- schedule_delayed_work(&jpeg->job_timeout_work,
- msecs_to_jiffies(MTK_JPEG_HW_TIMEOUT_MSEC));
-
mtk_jpeg_set_dec_src(ctx, &src_buf->vb2_buf, &bs);
if (mtk_jpeg_set_dec_dst(ctx, &jpeg_src_buf->dec_param, &dst_buf->vb2_buf, &fb))
goto dec_end;

+ schedule_delayed_work(&jpeg->job_timeout_work,
+ msecs_to_jiffies(MTK_JPEG_HW_TIMEOUT_MSEC));
+
spin_lock_irqsave(&jpeg->hw_lock, flags);
mtk_jpeg_dec_reset(jpeg->reg_base);
mtk_jpeg_dec_set_config(jpeg->reg_base,

--
Best regards,
Dmitry