Patchwork wireless extensions: play with netns

login
register
mail settings
Submitter Johannes Berg
Date June 18, 2009, 12:39 a.m.
Message ID <1245285541.31588.79.camel@johannes.local>
Download mbox | patch
Permalink /patch/28825/
State RFC
Delegated to: David Miller
Headers show

Comments

Johannes Berg - June 18, 2009, 12:39 a.m.
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@sipsolutions.net>
---
 include/net/net_namespace.h |    3 +++
 net/core/net_namespace.c    |   10 +++++++---
 2 files changed, 10 insertions(+), 3 deletions(-)



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Eric W. Biederman - June 18, 2009, 1:12 a.m.
Johannes Berg <johannes@sipsolutions.net> writes:

> 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.

Except for the changelog which is wrong about not needing any rcu
primitives I can't spot any bugs.

> 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@sipsolutions.net>
> ---
>  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 netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

--- 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);