2024-01-12 08:43:21

by Zhipeng Lu

[permalink] [raw]
Subject: [PATCH] media: atomisp: ssh_css: Fix a null-pointer dereference in load_video_binaries

The allocation failure of mycs->yuv_scaler_binary in load_video_binaries
is followed with a dereference of mycs->yuv_scaler_binary after the
following call chain:

sh_css_pipe_load_binaries
|-> load_video_binaries (mycs->yuv_scaler_binary == NULL)
|
|-> sh_css_pipe_unload_binaries
|-> unload_video_binaries

In unload_video_binaries, it calls to ia_css_binary_unload with argument
&pipe->pipe_settings.video.yuv_scaler_binary[i], which refers to the
same memory slot as mycs->yuv_scaler_binary. Thus, a null-pointer
dereference is triggered.

Fixes: ad85094b293e ("Revert "media: staging: atomisp: Remove driver"")
Signed-off-by: Zhipeng Lu <[email protected]>
---
drivers/staging/media/atomisp/pci/sh_css.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/staging/media/atomisp/pci/sh_css.c b/drivers/staging/media/atomisp/pci/sh_css.c
index f35c90809414..eb43f4e99d02 100644
--- a/drivers/staging/media/atomisp/pci/sh_css.c
+++ b/drivers/staging/media/atomisp/pci/sh_css.c
@@ -4936,9 +4936,10 @@ unload_video_binaries(struct ia_css_pipe *pipe)
ia_css_binary_unload(&pipe->pipe_settings.video.video_binary);
ia_css_binary_unload(&pipe->pipe_settings.video.vf_pp_binary);

- for (i = 0; i < pipe->pipe_settings.video.num_yuv_scaler; i++)
- ia_css_binary_unload(&pipe->pipe_settings.video.yuv_scaler_binary[i]);
-
+ if (pipe->pipe_settings.video.yuv_scaler_binary)
+ for (i = 0; i < pipe->pipe_settings.video.num_yuv_scaler; i++)
+ ia_css_binary_unload(&pipe->pipe_settings.video.yuv_scaler_binary[i]);
+
kfree(pipe->pipe_settings.video.is_output_stage);
pipe->pipe_settings.video.is_output_stage = NULL;
kfree(pipe->pipe_settings.video.yuv_scaler_binary);
--
2.34.1



2024-01-12 08:49:40

by Hans de Goede

[permalink] [raw]
Subject: Re: [PATCH] media: atomisp: ssh_css: Fix a null-pointer dereference in load_video_binaries

Hi Zhipeng Lu,

On 1/12/24 09:34, Zhipeng Lu wrote:
> The allocation failure of mycs->yuv_scaler_binary in load_video_binaries
> is followed with a dereference of mycs->yuv_scaler_binary after the
> following call chain:
>
> sh_css_pipe_load_binaries
> |-> load_video_binaries (mycs->yuv_scaler_binary == NULL)
> |
> |-> sh_css_pipe_unload_binaries
> |-> unload_video_binaries
>
> In unload_video_binaries, it calls to ia_css_binary_unload with argument
> &pipe->pipe_settings.video.yuv_scaler_binary[i], which refers to the
> same memory slot as mycs->yuv_scaler_binary. Thus, a null-pointer
> dereference is triggered.
>
> Fixes: ad85094b293e ("Revert "media: staging: atomisp: Remove driver"")
> Signed-off-by: Zhipeng Lu <[email protected]>

Thank you for your patch. I believe it would be better to fix this
like this:

diff --git a/drivers/staging/media/atomisp/pci/sh_css.c b/drivers/staging/media/atomisp/pci/sh_css.c
index 1d1fbda75da1..d566c5417448 100644
--- a/drivers/staging/media/atomisp/pci/sh_css.c
+++ b/drivers/staging/media/atomisp/pci/sh_css.c
@@ -4690,6 +4690,7 @@ static int load_video_binaries(struct ia_css_pipe *pipe)
sizeof(struct ia_css_binary),
GFP_KERNEL);
if (!mycs->yuv_scaler_binary) {
+ mycs->num_yuv_scaler = 0;
err = -ENOMEM;
return err;
}

Can you please submit a new version using this approach ?

Regards,

Hans



> ---
> drivers/staging/media/atomisp/pci/sh_css.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/staging/media/atomisp/pci/sh_css.c b/drivers/staging/media/atomisp/pci/sh_css.c
> index f35c90809414..eb43f4e99d02 100644
> --- a/drivers/staging/media/atomisp/pci/sh_css.c
> +++ b/drivers/staging/media/atomisp/pci/sh_css.c
> @@ -4936,9 +4936,10 @@ unload_video_binaries(struct ia_css_pipe *pipe)
> ia_css_binary_unload(&pipe->pipe_settings.video.video_binary);
> ia_css_binary_unload(&pipe->pipe_settings.video.vf_pp_binary);
>
> - for (i = 0; i < pipe->pipe_settings.video.num_yuv_scaler; i++)
> - ia_css_binary_unload(&pipe->pipe_settings.video.yuv_scaler_binary[i]);
> -
> + if (pipe->pipe_settings.video.yuv_scaler_binary)
> + for (i = 0; i < pipe->pipe_settings.video.num_yuv_scaler; i++)
> + ia_css_binary_unload(&pipe->pipe_settings.video.yuv_scaler_binary[i]);
> +
> kfree(pipe->pipe_settings.video.is_output_stage);
> pipe->pipe_settings.video.is_output_stage = NULL;
> kfree(pipe->pipe_settings.video.yuv_scaler_binary);


2024-01-12 18:28:41

by Andy Shevchenko

[permalink] [raw]
Subject: Re: [PATCH] media: atomisp: ssh_css: Fix a null-pointer dereference in load_video_binaries

On Fri, Jan 12, 2024 at 10:49 AM Hans de Goede <[email protected]> wrote:
> On 1/12/24 09:34, Zhipeng Lu wrote:


> > Fixes: ad85094b293e ("Revert "media: staging: atomisp: Remove driver"")

> Can you please submit a new version using this approach ?

Besides that, are you sure the Fixes refers to the correct commit?

--
With Best Regards,
Andy Shevchenko

2024-01-17 20:51:14

by Andy Shevchenko

[permalink] [raw]
Subject: Re: [PATCH] media: atomisp: ssh_css: Fix a null-pointer dereference in load_video_binaries

On Wed, Jan 17, 2024 at 8:34 AM <[email protected]> wrote:
> > On Fri, Jan 12, 2024 at 10:49 AM Hans de Goede <hdegoede@redhatcom> wrote:
> > > On 1/12/24 09:34, Zhipeng Lu wrote:

> > > > Fixes: ad85094b293e ("Revert "media: staging: atomisp: Remove driver"")
> > Besides that, are you sure the Fixes refers to the correct commit?
>
> Well, I think I referred to the correct commit, which introduce the whole module and leave this bug.
>
> If I did miss something please let me know.

Yes, the driver was before that commit in the kernel. Was it without
the bug? No, because you are referring to a clear revert. So, find the
real commit that had brought that into the kernel.

--
With Best Regards,
Andy Shevchenko

2024-01-18 15:12:10

by Zhipeng Lu

[permalink] [raw]
Subject: Re: [PATCH] media: atomisp: ssh_css: Fix a null-pointer dereference in load_video_binaries

> On Wed, Jan 17, 2024 at 8:34 AM <[email protected]> wrote:
> > > On Fri, Jan 12, 2024 at 10:49 AM Hans de Goede <[email protected]> wrote:
> > > > On 1/12/24 09:34, Zhipeng Lu wrote:
>
> > > > > Fixes: ad85094b293e ("Revert "media: staging: atomisp: Remove driver"")
> > > Besides that, are you sure the Fixes refers to the correct commit?
> >
> > Well, I think I referred to the correct commit, which introduce the whole module and leave this bug.
> >
> > If I did miss something please let me know.
>
> Yes, the driver was before that commit in the kernel. Was it without
> the bug? No, because you are referring to a clear revert. So, find the
> real commit that had brought that into the kernel.

You are correct, I just did some git blame on the latest version but
forgot about the commit blamed was a revert commit.
A v2 version of this patch will be sent later to fix this issue.
Thank you for pointing out my mistake.

Regards,
Zhipeng

>
> --
> With Best Regards,
> Andy Shevchenko