Merge branch 'netdev_unregister_races'

Julian Anastasov says:

====================
net: fixes for device unregistration

Test script from Eric W. Biederman can catch a problem
where packets from backlog are processed long after the last
synchronize_net call. This can be reproduced after few tests
if commit 381c759d9916 ("ipv4: Avoid crashing in ip_error")
is reverted for the test. Incoming packets do not hold
reference to device but even if they do, subsystems do not
expect packets to fly during and after the NETDEV_UNREGISTER
event.

The first fix has the cost of netif_running check in fast path.
The second fix calls rcu_read_lock while local IRQ is disabled,
I hope this is not against the rules.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

net/core/dev.c

@@ -3448,6 +3448,8 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
 	local_irq_save(flags);
 
 	rps_lock(sd);
+	if (!netif_running(skb->dev))
+		goto drop;
 	qlen = skb_queue_len(&sd->input_pkt_queue);
 	if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
 		if (qlen) {
@@ -3469,6 +3471,7 @@ enqueue:
 		goto enqueue;
 	}
 
+drop:
 	sd->dropped++;
 	rps_unlock(sd);
 
@@ -3771,8 +3774,6 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
 
 	pt_prev = NULL;
 
-	rcu_read_lock();
-
 another_round:
 	skb->skb_iif = skb->dev->ifindex;
 
@@ -3782,7 +3783,7 @@ another_round:
 	    skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
 		skb = skb_vlan_untag(skb);
 		if (unlikely(!skb))
-			goto unlock;
+			goto out;
 	}
 
 #ifdef CONFIG_NET_CLS_ACT
@@ -3812,10 +3813,10 @@ skip_taps:
 	if (static_key_false(&ingress_needed)) {
 		skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
 		if (!skb)
-			goto unlock;
+			goto out;
 
 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
-			goto unlock;
+			goto out;
 	}
 #endif
 #ifdef CONFIG_NET_CLS_ACT
@@ -3833,7 +3834,7 @@ ncls:
 		if (vlan_do_receive(&skb))
 			goto another_round;
 		else if (unlikely(!skb))
-			goto unlock;
+			goto out;
 	}
 
 	rx_handler = rcu_dereference(skb->dev->rx_handler);
@@ -3845,7 +3846,7 @@ ncls:
 		switch (rx_handler(&skb)) {
 		case RX_HANDLER_CONSUMED:
 			ret = NET_RX_SUCCESS;
-			goto unlock;
+			goto out;
 		case RX_HANDLER_ANOTHER:
 			goto another_round;
 		case RX_HANDLER_EXACT:
@@ -3899,8 +3900,7 @@ drop:
 		ret = NET_RX_DROP;
 	}
 
-unlock:
-	rcu_read_unlock();
+out:
 	return ret;
 }
 
@@ -3931,29 +3931,30 @@ static int __netif_receive_skb(struct sk_buff *skb)
 
 static int netif_receive_skb_internal(struct sk_buff *skb)
 {
+	int ret;
+
 	net_timestamp_check(netdev_tstamp_prequeue, skb);
 
 	if (skb_defer_rx_timestamp(skb))
 		return NET_RX_SUCCESS;
 
+	rcu_read_lock();
+
 #ifdef CONFIG_RPS
 	if (static_key_false(&rps_needed)) {
 		struct rps_dev_flow voidflow, *rflow = &voidflow;
-		int cpu, ret;
-
-		rcu_read_lock();
-
-		cpu = get_rps_cpu(skb->dev, skb, &rflow);
+		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
 
 		if (cpu >= 0) {
 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
 			rcu_read_unlock();
 			return ret;
 		}
-		rcu_read_unlock();
 	}
 #endif
-	return __netif_receive_skb(skb);
+	ret = __netif_receive_skb(skb);
+	rcu_read_unlock();
+	return ret;
 }
 
 /**
@@ -4498,8 +4499,10 @@ static int process_backlog(struct napi_struct *napi, int quota)
 		struct sk_buff *skb;
 
 		while ((skb = __skb_dequeue(&sd->process_queue))) {
+			rcu_read_lock();
 			local_irq_enable();
 			__netif_receive_skb(skb);
+			rcu_read_unlock();
 			local_irq_disable();
 			input_queue_head_incr(sd);
 			if (++work >= quota) {
@@ -6135,6 +6138,7 @@ static void rollback_registered_many(struct list_head *head)
 		unlist_netdevice(dev);
 
 		dev->reg_state = NETREG_UNREGISTERING;
+		on_each_cpu(flush_backlog, dev, 1);
 	}
 
 	synchronize_net();
@@ -6770,8 +6774,6 @@ void netdev_run_todo(void)
 
 		dev->reg_state = NETREG_UNREGISTERED;
 
-		on_each_cpu(flush_backlog, dev, 1);
-
 		netdev_wait_allrefs(dev);
 
 		/* paranoia */