net: avoid theoretical races with ref drain

Technically, it's illegal to take a ref on a netdev just because
we have a pointer on which we already hold a ref, with no other
protection. This is because our simple per-cpu refcount
implementation cannot atomically read the count.

Let's make sure we cancel outstanding work and never queue more
work for a device we know is dead. This way taking a ref on
a dev we know is on the netdev_work_list is always going to be safe.

Jiangshan Yi reports that the issues is caught by ref tracker infra
leading to a warning:
  WARNING: lib/ref_tracker.c:322 at ref_tracker_free
  WARNING: lib/ref_tracker.c:246 at ref_tracker_dir_exit

Reported-by: Jiangshan Yi <[email protected]>
Link: https://lore.kernel.org/[email protected]
Fixes: 12c765be84 ("net: turn the rx_mode work into a generic netdev_work facility")
Link: https://patch.msgid.link/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net/core/dev.c

@@ -12436,6 +12436,7 @@ void unregister_netdevice_many_notify(struct list_head *head,
 		dev_tcx_uninstall(dev);
 		dev_xdp_uninstall(dev);
 		dev_memory_provider_uninstall(dev);
+		netdev_work_cancel_all(dev);
 		netdev_unlock_ops(dev);
 		bpf_dev_bound_netdev_unregister(dev);
 

net/core/dev.h

@@ -179,6 +179,7 @@ enum netdev_work_core {
 void __netdev_work_core_sched(struct net_device *dev, unsigned long event);
 unsigned long
 __netdev_work_core_cancel(struct net_device *dev, unsigned long mask);
+void netdev_work_cancel_all(struct net_device *dev);
 
 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
 			unsigned int gchanges, u32 portid,

net/core/netdev_work.c

@@ -31,6 +31,10 @@ static void netdev_work_enqueue(struct net_device *dev, unsigned long events,
 		return;
 
 	spin_lock_bh(&netdev_work_lock);
+	if (!dev_isalive(dev)) {
+		spin_unlock_bh(&netdev_work_lock);
+		return;
+	}
 	if (list_empty(&dev->work_node)) {
 		list_add_tail(&dev->work_node, &netdev_work_list);
 		netdev_hold(dev, &dev->work_tracker, GFP_ATOMIC);
@@ -61,6 +65,18 @@ netdev_work_dequeue(struct net_device *dev, unsigned long *pending,
 	return events;
 }
 
+void netdev_work_cancel_all(struct net_device *dev)
+{
+	spin_lock_bh(&netdev_work_lock);
+	dev->work_pending = 0;
+	dev->work_core_pending = 0;
+	if (!list_empty(&dev->work_node)) {
+		list_del_init(&dev->work_node);
+		netdev_put(dev, &dev->work_tracker);
+	}
+	spin_unlock_bh(&netdev_work_lock);
+}
+
 void netdev_work_sched(struct net_device *dev, unsigned long events)
 {
 	netdev_work_enqueue(dev, events, 0);