net: remove delay at device dismantle
I noticed extra one second delay in device dismantle, tracked down to
a call to dst_dev_event() while some call_rcu() are still in RCU queues.
These call_rcu() were posted by rt_free(struct rtable *rt) calls.
We then wait a little (but one second) in netdev_wait_allrefs() before
kicking again NETDEV_UNREGISTER.
As the call_rcu() are now completed, dst_dev_event() can do the needed
device swap on busy dst.
To solve this problem, add a new NETDEV_UNREGISTER_FINAL, called
after a rcu_barrier(), but outside of RTNL lock.
Use NETDEV_UNREGISTER_FINAL with care !
Change dst_dev_event() handler to react to NETDEV_UNREGISTER_FINAL
Also remove NETDEV_UNREGISTER_BATCH, as its not used anymore after
IP cache removal.
With help from Gao feng
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/core/dev.c b/net/core/dev.c
index 088923f..0640d2a 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1406,7 +1406,6 @@
nb->notifier_call(nb, NETDEV_DOWN, dev);
}
nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
- nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
}
}
@@ -1448,7 +1447,6 @@
nb->notifier_call(nb, NETDEV_DOWN, dev);
}
nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
- nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
}
}
unlock:
@@ -1468,7 +1466,8 @@
int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
{
- ASSERT_RTNL();
+ if (val != NETDEV_UNREGISTER_FINAL)
+ ASSERT_RTNL();
return raw_notifier_call_chain(&netdev_chain, val, dev);
}
EXPORT_SYMBOL(call_netdevice_notifiers);
@@ -5331,10 +5330,6 @@
netdev_unregister_kobject(dev);
}
- /* Process any work delayed until the end of the batch */
- dev = list_first_entry(head, struct net_device, unreg_list);
- call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
-
synchronize_net();
list_for_each_entry(dev, head, unreg_list)
@@ -5787,9 +5782,8 @@
/* Rebroadcast unregister notification */
call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
- /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
- * should have already handle it the first time */
-
+ rcu_barrier();
+ call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
&dev->state)) {
/* We must not have linkwatch events
@@ -5851,9 +5845,8 @@
__rtnl_unlock();
- /* Wait for rcu callbacks to finish before attempting to drain
- * the device list. This usually avoids a 250ms wait.
- */
+
+ /* Wait for rcu callbacks to finish before next phase */
if (!list_empty(&list))
rcu_barrier();
@@ -5862,6 +5855,8 @@
= list_first_entry(&list, struct net_device, todo_list);
list_del(&dev->todo_list);
+ call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
+
if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
pr_err("network todo '%s' but state %d\n",
dev->name, dev->reg_state);
@@ -6256,7 +6251,6 @@
the device is just moving and can keep their slaves up.
*/
call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
- call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
/*
