On Wed, 2009-06-17 at 17:30 -0700, Eric W. Biederman wrote:
> >> This of course gives you a network namespace that can be found by for_each_net rcu
> >> while the per net exit functions are running. I think that opens up to races
> >> that I don't want to think about.
> >
> > Indeed. Can we move the rcu_barrier() up? Or we could insert a
> > synchronize_rcu() (which is sufficient for rcu_read_lock) before the
> > exit functions are run?
>
> I don't think we can move the barrier. But adding an extra synchronize_rcu
> should be fine. We are talking about the slowest of the slow paths here.
> It is so slow it even gets it's own kernel thread.
Ok :) New patch below.
Yes, you're right, we can't move it, it might still be used from exit
for net_assign_generic or so.
johannes
Subject: net: make namespace iteration possible under RCU
We already call rcu_barrier(), so all we need to take
care of is using proper RCU list add/del primitives.
Signed-off-by: Johannes Berg <johannes@xxxxxxxxxxxxxxxx>
---
include/net/net_namespace.h | 3 +++
net/core/net_namespace.c | 10 +++++++---
2 files changed, 10 insertions(+), 3 deletions(-)
--- wireless-testing.orig/include/net/net_namespace.h 2009-06-18 01:36:26.000000000 +0200
+++ wireless-testing/include/net/net_namespace.h 2009-06-18 02:17:14.000000000 +0200
@@ -211,6 +211,9 @@ static inline struct net *read_pnet(stru
#define for_each_net(VAR) \
list_for_each_entry(VAR, &net_namespace_list, list)
+#define for_each_net_rcu(VAR) \
+ list_for_each_entry_rcu(VAR, &net_namespace_list, list)
+
#ifdef CONFIG_NET_NS
#define __net_init
#define __net_exit
--- wireless-testing.orig/net/core/net_namespace.c 2009-06-18 01:36:39.000000000 +0200
+++ wireless-testing/net/core/net_namespace.c 2009-06-18 02:03:06.000000000 +0200
@@ -6,6 +6,7 @@
#include <linux/delay.h>
#include <linux/sched.h>
#include <linux/idr.h>
+#include <linux/rculist.h>
#include <net/net_namespace.h>
#include <net/netns/generic.h>
@@ -134,7 +135,7 @@ struct net *copy_net_ns(unsigned long fl
err = setup_net(new_net);
if (!err) {
rtnl_lock();
- list_add_tail(&new_net->list, &net_namespace_list);
+ list_add_tail_rcu(&new_net->list, &net_namespace_list);
rtnl_unlock();
}
mutex_unlock(&net_mutex);
@@ -163,9 +164,12 @@ static void cleanup_net(struct work_stru
/* Don't let anyone else find us. */
rtnl_lock();
- list_del(&net->list);
+ list_del_rcu(&net->list);
rtnl_unlock();
+ /* if somebody is rcu-iterating the list, wait */
+ synchronize_rcu();
+
/* Run all of the network namespace exit methods */
list_for_each_entry_reverse(ops, &pernet_list, list) {
if (ops->exit)
@@ -227,7 +231,7 @@ static int __init net_ns_init(void)
err = setup_net(&init_net);
rtnl_lock();
- list_add_tail(&init_net.list, &net_namespace_list);
+ list_add_tail_rcu(&init_net.list, &net_namespace_list);
rtnl_unlock();
mutex_unlock(&net_mutex);
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
