Re: [PATCH] drm/xe/guc_submit: fix UAF in run_job()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Sep 20, 2024 at 01:38:07PM +0100, Matthew Auld wrote:
> The initial kref from dma_fence_init() should match up with whatever
> signals the fence, however here we are submitting the job first to the
> hw and only then grabbing the extra ref and even then we touch some
> fence state before this. This might be too late if the fence is
> signalled before we can grab the extra ref. Rather always grab the
> refcount early before we do the submission part.
> 

I think I see the race. Let me make sure I understand.

Current flow:

1. guc_exec_queue_run_job enters
2. guc_exec_queue_run_job submits job to hardware
3. job finishes on hardware
4. irq handler for job completion fires, signals job->fence, does last
   put on job->fence freeing the memory
5. guc_exec_queue_run_job takes a ref job->fence and BOOM UAF

The extra ref between steps 1/2 dropped after 5 prevents this. Is that
right?

Assuming my understanding is correct:
Reviewed-by: Matthew Brost <matthew.brost@xxxxxxxxx>

> Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2811
> Signed-off-by: Matthew Auld <matthew.auld@xxxxxxxxx>
> Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
> Cc: Matthew Brost <matthew.brost@xxxxxxxxx>
> Cc: <stable@xxxxxxxxxxxxxxx> # v6.8+
> ---
>  drivers/gpu/drm/xe/xe_guc_submit.c | 13 ++++++++++---
>  1 file changed, 10 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
> index fbbe6a487bbb..b33f3d23a068 100644
> --- a/drivers/gpu/drm/xe/xe_guc_submit.c
> +++ b/drivers/gpu/drm/xe/xe_guc_submit.c
> @@ -766,12 +766,15 @@ guc_exec_queue_run_job(struct drm_sched_job *drm_job)
>  	struct xe_guc *guc = exec_queue_to_guc(q);
>  	struct xe_device *xe = guc_to_xe(guc);
>  	bool lr = xe_exec_queue_is_lr(q);
> +	struct dma_fence *fence;
>  
>  	xe_assert(xe, !(exec_queue_destroyed(q) || exec_queue_pending_disable(q)) ||
>  		  exec_queue_banned(q) || exec_queue_suspended(q));
>  
>  	trace_xe_sched_job_run(job);
>  
> +	dma_fence_get(job->fence);
> +
>  	if (!exec_queue_killed_or_banned_or_wedged(q) && !xe_sched_job_is_error(job)) {
>  		if (!exec_queue_registered(q))
>  			register_exec_queue(q);
> @@ -782,12 +785,16 @@ guc_exec_queue_run_job(struct drm_sched_job *drm_job)
>  
>  	if (lr) {
>  		xe_sched_job_set_error(job, -EOPNOTSUPP);
> -		return NULL;
> +		fence = NULL;
>  	} else if (test_and_set_bit(JOB_FLAG_SUBMIT, &job->fence->flags)) {
> -		return job->fence;
> +		fence = job->fence;
>  	} else {
> -		return dma_fence_get(job->fence);
> +		fence = dma_fence_get(job->fence);
>  	}
> +
> +	dma_fence_put(job->fence);
> +
> +	return fence;
>  }
>  
>  static void guc_exec_queue_free_job(struct drm_sched_job *drm_job)
> -- 
> 2.46.0
> 




[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux