[PATCH] Prevent "eth0: driver changed get_stats after register" lying

[PATCH] Prevent "eth0: driver changed get_stats after register" lying

[Russell King]

Hi,

There appears to be a race condition between netdev_register_sysfs
and hotplug.  I'm seeing the message in the subject upon initialisation
of pcnet_cs.

We set "dev->last_stats" after we have registered the class device.
During class device registration, we trigger the hotplug scripts,
which can in turn bring up the interface.  This in turn checks to
see if "dev->last_stats == dev->get_stats", and, since we haven't
set dev->last_stats yet, we complain.

Here is a patch against 2.6.0-test8 which moves the initialisation
of last_stats such that we avoid the race.

--- orig/net/core/net-sysfs.c	Sat Oct 18 00:00:38 2003
+++ linux/net/core/net-sysfs.c	Sat Oct 18 21:23:15 2003
@@ -408,6 +408,7 @@
 
 	class_dev->class = &net_class;
 	class_dev->class_data = net;
+	net->last_stats = net->get_stats;
 
 	strlcpy(class_dev->class_id, net->name, BUS_ID_SIZE);
 	if ((ret = class_device_register(class_dev)))
@@ -419,7 +420,6 @@
 	}
 
 
-	net->last_stats = net->get_stats;
 	if (net->get_stats &&
 	    (ret = sysfs_create_group(&class_dev->kobj, &netstat_group)))
 		goto out_unreg; 


-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core

Re: [PATCH] Prevent "eth0: driver changed get_stats after register" lying

[Russell King]

On Sat, Oct 18, 2003 at 09:46:22PM +0100, Russell King wrote:
> There appears to be a race condition between netdev_register_sysfs
> and hotplug.  I'm seeing the message in the subject upon initialisation
> of pcnet_cs.
> 
> We set "dev->last_stats" after we have registered the class device.
> During class device registration, we trigger the hotplug scripts,
> which can in turn bring up the interface.  This in turn checks to
> see if "dev->last_stats == dev->get_stats", and, since we haven't
> set dev->last_stats yet, we complain.

Incidentally, this also means that hotplug scripts can run before
we've registered the class device attributes.  Could that be another
problem?  It's probably something that the hotplug people should
be aware of.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core

Re: [PATCH] Prevent "eth0: driver changed get_stats after register" lying

[David S. Miller]

On Sat, 18 Oct 2003 21:46:22 +0100
Russell King <rmk@arm.linux.org.uk> wrote:

> We set "dev->last_stats" after we have registered the class device.
> During class device registration, we trigger the hotplug scripts,
> which can in turn bring up the interface.  This in turn checks to
> see if "dev->last_stats == dev->get_stats", and, since we haven't
> set dev->last_stats yet, we complain.
> 
> Here is a patch against 2.6.0-test8 which moves the initialisation
> of last_stats such that we avoid the race.

Looks ok, thanks Russell.