bugfix: Mark m_tip_block_cv as guarded by m_tip_block_mutex

This is not strictly required, but all places using m_tip_block_cv
(except shutdown) already take the lock. The annotation makes it easier
to catch potential deadlocks before review.

Adding the missing lock to the shutdown sequence is a bugfix.

An alternative would be to take the lock and release it before
notifying, see
bitcoin#30967 (comment)

Author: MarcoFalke

src/node/kernel_notifications.h

@@ -58,7 +58,7 @@ class KernelNotifications : public kernel::Notifications
     bool m_shutdown_on_fatal_error{true};
 
     Mutex m_tip_block_mutex;
-    std::condition_variable m_tip_block_cv;
+    std::condition_variable m_tip_block_cv GUARDED_BY(m_tip_block_mutex);
     //! The block for which the last blockTip notification was received for.
     uint256 m_tip_block GUARDED_BY(m_tip_block_mutex);
 

src/rpc/server.cpp

@@ -305,7 +305,7 @@ void StopRPC(const std::any& context)
         DeleteAuthCookie();
         node::NodeContext& node = EnsureAnyNodeContext(context);
         // The notifications interface doesn't exist between initialization step 4a and 7.
-        if (node.notifications) node.notifications->m_tip_block_cv.notify_all();
+        if (node.notifications) WITH_LOCK(node.notifications->m_tip_block_mutex, node.notifications->m_tip_block_cv.notify_all());
         LogDebug(BCLog::RPC, "RPC stopped.\n");
     });
 }