Thank you for your contribution! Sashiko AI review found no new issues in this patch, but noticed 1 related pre-existing issue it does not introduce: - [Critical] rust: dma_fence: missing RCU grace period in DriverFence::drop() [1] --
commit ee6d42825838039a4b13a90afb76009d7c94efbd Author: Philipp Stanner <[email protected]> rust: DmaFence: Don't drop on signal This commit changes DriverFence::signal() to take &self instead of consuming self, and decrements nr_of_unsignaled_fences only when the fence had not yet been signaled. > diff --git a/rust/kernel/dma_buf/dma_fence.rs > b/rust/kernel/dma_buf/dma_fence.rs > index 897308f60dd3c..c04b68fe67fe3 100644 > --- a/rust/kernel/dma_buf/dma_fence.rs > +++ b/rust/kernel/dma_buf/dma_fence.rs > @@ -820,7 +820,7 @@ pub fn as_fence(&self) -> &Fence { > } > > /// Signal the fence. This will invoke all registered callbacks. > - pub fn signal(self, res: Result) { > + pub fn signal(&self, res: Result) { > let fence = self.as_fence().lock(); > > // SAFETY: `fence` is valid because `self` is valid. The lock must be > @@ -831,14 +831,15 @@ pub fn signal(self, res: Result) { > // must not have been signaled yet, which we check directly > above. > unsafe { bindings::dma_fence_set_error(fence.as_raw(), > err.to_errno()) }; > } > + > + // SAFETY: `self.data` is valid because `self` is valid. > + let fctx = unsafe { self.data.as_ref().fctx }; > + let _ = fctx.nr_of_unsignaled_fences.fetch_sub(1, Relaxed); > + > // SAFETY: `fence` is valid because `self` is valid. The lock > must > // be held, which we acquired above. > unsafe { bindings::dma_fence_signal_locked(fence.as_raw()) }; > } > - > - // SAFETY: `self.data` is valid because `self` is valid. > - let fctx = unsafe { self.data.as_ref().fctx }; > - let _ = fctx.nr_of_unsignaled_fences.fetch_sub(1, Relaxed); > } > } [Severity: Critical] This problem wasn't introduced by this patch, but can FenceContext::drop() free the context before an RCU grace period has elapsed [1] when T::FenceDataType does not require drop glue? FenceContext::drop() calls rcu_barrier() to wait for in-flight RCU callbacks and concurrent dma_fence_ops readers before freeing the FenceContext. However, in DriverFence::drop(): rust/kernel/dma_buf/dma_fence.rs:DriverFence::drop() { ... if !core::mem::needs_drop::<T::FenceDataType>() { ... unsafe { bindings::dma_fence_put(self.as_raw()) }; return; } ... unsafe { bindings::call_rcu(rcu_head_ptr, Some(drop_driver_fence_data::<T>)) }; } When needs_drop is false and an external consumer still holds a reference to the underlying dma_fence, dma_fence_put() does not schedule the RCU release of the fence and DriverFence::drop() returns without calling call_rcu(). Because rcu_barrier() only waits for already-queued RCU callbacks rather than a grace period, FenceContext::drop() can return immediately while a concurrent RCU reader in dma_fence_driver_name() or dma_fence_timeline_name() is still running: CPU0 CPU1 dma_fence_driver_name() rcu_read_lock() !dma_fence_test_signaled_flag() DriverFence::signal() nr_of_unsignaled_fences.fetch_sub(1) dma_fence_signal_locked() DriverFence::drop() // !needs_drop::<T::FenceDataType>() dma_fence_put() // refcount > 0 FenceContext::drop() rcu_barrier() // no callback queued // FenceContext is freed FenceContext::get_driver_name() fctx.driver_name.as_char_ptr() <-- use-after-free [1] https://sashiko.dev/#/bug/linux-40274ecc-9f61-4d35-b6b3-1ceb72f48227 -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=3
