2007-06-08 07:27:13

by Chris Wright

[permalink] [raw]
Subject: [patch 29/32] NET: Fix race condition about network device name allocation.

-stable review patch. If anyone has any objections, please let us know.
---------------------

From: Stephen Hemminger <[email protected]>

Kenji Kaneshige found this race between device removal and
registration. On unregister it is possible for the old device to
exist, because sysfs file is still open. A new device with 'eth%d'
will select the same name, but sysfs kobject register will fial.

The following changes the shutdown order slightly. It hold a removes
the sysfs entries earlier (on unregister_netdevice), but holds a
kobject reference. Then when todo runs the actual last put free
happens.

Signed-off-by: Stephen Hemminger <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
[chrisw: backport to 2.6.20]
Signed-off-by: Chris Wright <[email protected]>

---
net/core/dev.c | 10 ++++++----
net/core/net-sysfs.c | 8 +++++++-
2 files changed, 13 insertions(+), 5 deletions(-)

--- linux-2.6.20.13.orig/net/core/dev.c
+++ linux-2.6.20.13/net/core/dev.c
@@ -3138,7 +3138,6 @@ void netdev_run_todo(void)
continue;
}

- netdev_unregister_sysfs(dev);
dev->reg_state = NETREG_UNREGISTERED;

netdev_wait_allrefs(dev);
@@ -3149,11 +3148,11 @@ void netdev_run_todo(void)
BUG_TRAP(!dev->ip6_ptr);
BUG_TRAP(!dev->dn_ptr);

- /* It must be the very last action,
- * after this 'dev' may point to freed up memory.
- */
if (dev->destructor)
dev->destructor(dev);
+
+ /* Free network device */
+ kobject_put(&dev->dev.kobj);
}

out:
@@ -3310,6 +3309,9 @@ int unregister_netdevice(struct net_devi
/* Notifier chain MUST detach us from master device. */
BUG_TRAP(!dev->master);

+ /* Remove entries from sysfs */
+ netdev_unregister_sysfs(dev);
+
/* Finish processing unregister after unlock */
net_set_todo(dev);

--- linux-2.6.20.13.orig/net/core/net-sysfs.c
+++ linux-2.6.20.13/net/core/net-sysfs.c
@@ -440,9 +440,15 @@ static struct class net_class = {
#endif
};

+/* Delete sysfs entries but hold kobject reference until after all
+ * netdev references are gone.
+ */
void netdev_unregister_sysfs(struct net_device * net)
{
- class_device_del(&(net->class_dev));
+ struct device *dev = &(net->dev);
+
+ kobject_get(&dev->kobj);
+ device_del(dev);
}

/* Create sysfs entries for network device. */

--


2007-06-08 14:35:13

by koan

[permalink] [raw]
Subject: RE: [patch 29/32] NET: Fix race condition about network device name allocation.

Hello,

I get a compilation failure from the patch:

CC net/core/stream.o
CC net/core/scm.o
CC net/core/gen_stats.o
CC net/core/gen_estimator.o
CC net/core/sysctl_net_core.o
CC net/core/dev.o
net/core/dev.c: In function 'netdev_run_todo':
net/core/dev.c:3155: error: 'struct net_device' has no member named 'dev'
make[2]: *** [net/core/dev.o] Error 1
make[1]: *** [net/core] Error 2
make: *** [net] Error 2


Thanks, Jesse

(I am not on list, please CC me with replies)

2007-06-08 16:34:36

by Chris Wright

[permalink] [raw]
Subject: Re: [patch 29/32] NET: Fix race condition about network device name allocation.

* koan ([email protected]) wrote:
> I get a compilation failure from the patch:
>
> CC net/core/stream.o
> CC net/core/scm.o
> CC net/core/gen_stats.o
> CC net/core/gen_estimator.o
> CC net/core/sysctl_net_core.o
> CC net/core/dev.o
> net/core/dev.c: In function 'netdev_run_todo':
> net/core/dev.c:3155: error: 'struct net_device' has no member named 'dev'
> make[2]: *** [net/core/dev.o] Error 1
> make[1]: *** [net/core] Error 2
> make: *** [net] Error 2

Yes, that's a terrible backport from me. I'm testing this one now,
and may simply drop it.
--

From: Stephen Hemminger <[email protected]>

Kenji Kaneshige found this race between device removal and
registration. On unregister it is possible for the old device to
exist, because sysfs file is still open. A new device with 'eth%d'
will select the same name, but sysfs kobject register will fial.

The following changes the shutdown order slightly. It hold a removes
the sysfs entries earlier (on unregister_netdevice), but holds a
kobject reference. Then when todo runs the actual last put free
happens.

Signed-off-by: Stephen Hemminger <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
[chrisw: backport to 2.6.20]
Signed-off-by: Chris Wright <[email protected]>

---
net/core/dev.c | 10 ++++++----
net/core/net-sysfs.c | 8 +++++++-
2 files changed, 13 insertions(+), 5 deletions(-)

--- linux-2.6.20.13.orig/net/core/dev.c
+++ linux-2.6.20.13/net/core/dev.c
@@ -3138,7 +3138,6 @@ void netdev_run_todo(void)
continue;
}

- netdev_unregister_sysfs(dev);
dev->reg_state = NETREG_UNREGISTERED;

netdev_wait_allrefs(dev);
@@ -3149,11 +3148,11 @@ void netdev_run_todo(void)
BUG_TRAP(!dev->ip6_ptr);
BUG_TRAP(!dev->dn_ptr);

- /* It must be the very last action,
- * after this 'dev' may point to freed up memory.
- */
if (dev->destructor)
dev->destructor(dev);
+
+ /* Free network device */
+ kobject_put(&dev->class_dev.kobj);
}

out:
@@ -3310,6 +3309,9 @@ int unregister_netdevice(struct net_devi
/* Notifier chain MUST detach us from master device. */
BUG_TRAP(!dev->master);

+ /* Remove entries from sysfs */
+ netdev_unregister_sysfs(dev);
+
/* Finish processing unregister after unlock */
net_set_todo(dev);

--- linux-2.6.20.13.orig/net/core/net-sysfs.c
+++ linux-2.6.20.13/net/core/net-sysfs.c
@@ -440,9 +440,15 @@ static struct class net_class = {
#endif
};

+/* Delete sysfs entries but hold kobject reference until after all
+ * netdev references are gone.
+ */
void netdev_unregister_sysfs(struct net_device * net)
{
- class_device_del(&(net->class_dev));
+ struct class_device *dev = &(net->class_dev);
+
+ kobject_get(&dev->kobj);
+ class_device_del(dev);
}

/* Create sysfs entries for network device. */