Hi Shaobo,

First of all, could you please make sure you send future mails to the linux-
media mailing list in plain text only (no HTML) ? The mailing list server 
rejects HTML e-mails.

On Thursday 16 Feb 2017 16:08:25 Shaobo wrote:
> Hi there,
> 
> My name is Shaobo He and I am a graduate student at University of Utah. I am
> applying a static analysis tool to the Linux device drivers, looking for
> NULL pointer dereference and accidentally found a plausible dead code
> location in v4l2-mem2mem.c due to undefined behavior.
> 
> The following is the problematic code segment,
> 
> static struct v4l2_m2m_queue_ctx *get_queue_ctx(struct v4l2_m2m_ctx
> *m2m_ctx,
>                                                 enum v4l2_buf_type type)
> {
>       if (V4L2_TYPE_IS_OUTPUT(type))
>               return &m2m_ctx->out_q_ctx;
>       else
>               return &m2m_ctx->cap_q_ctx;
> }
> 
> struct vb2_queue *v4l2_m2m_get_vq(struct v4l2_m2m_ctx *m2m_ctx,
>                                   enum v4l2_buf_type type)
> {
>       struct v4l2_m2m_queue_ctx *q_ctx;
> 
>       q_ctx = get_queue_ctx(m2m_ctx, type);
>       if (!q_ctx)
>               return NULL;
> 
>       return &q_ctx->q;
> }
> 
> `get_queue_ctx` returns a pointer value that is an addition of the base
> pointer address (`m2m_ctx`) to a non-zero offset. The following is the
> definition of struct v4l2_m2m_ctx,
> 
> struct v4l2_m2m_ctx {
>       /* optional cap/out vb2 queues lock */
>       struct mutex                    *q_lock;
> 
>       /* internal use only */
>       struct v4l2_m2m_dev             *m2m_dev;
> 
>       struct v4l2_m2m_queue_ctx       cap_q_ctx;
> 
>       struct v4l2_m2m_queue_ctx       out_q_ctx;
> 
>       /* For device job queue */
>       struct list_head                queue;
>       unsigned long                   job_flags;
>       wait_queue_head_t               finished;
> 
>       void                            *priv;
> };
> 
> There is a NULL test in a caller of `get_queue_ctx` (line 85), which appears
> problematic to me. I'm not sure if it is defined or feasible under the
> context of Linux kernel. This blog
> (https://wdtz.org/undefined-behavior-in-binutils-causes-segfault.html)
> suggests that the NULL check can be optimized away because the only case
> that the return value can be NULL triggers pointer overflow, which is
> undefined.
> 
> Please let me know if it makes sense or not. Thanks for your time and I am
> looking forward to your reply.

The NULL check is indeed wrong. I believe that the m2m_ctx argument passed to 
the v4l2_m2m_get_vq() function should never be NULL. We will however need to 
audit drivers to make sure that's the case. The NULL check could then be 
removed. Alternatively we could check m2m_ctx above the get_queue_ctx() call, 
which wouldn't require auditing drivers. It's a safe option, but would likely 
result in an unneeded NULL check.

-- 
Regards,

Laurent Pinchart

Reply via email to