Re: [PATCH net-next] sunvnet: call unregister_netdev() exactly once per device

Re: [PATCH net-next] sunvnet: call unregister_netdev() exactly once per device

[David Miller]

From: sowmini.varadhan@oracle.com
Date: Tue, 8 Jul 2014 12:18:27 -0400

> unregister_netdev the parent net_device only after
> the last port in the port_list of a `struct vnet'
> is removed when the vnet is dismantled. Moreover,
> this should only be done if the net_device was
> successfully registered.
> 
> Without this fix, 'modprobe -r sunvnet' could
> trigger the following in rollback_registered_many()
> 5693   BUG_ON(dev->reg_state != NETREG_REGISTERED);
> 
> Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
> Acked-by: Dave Kleikamp <dave.kleikamp@oracle.com>
> Acked-by: Dan Carpenter <dan.carpenter@oracle.com>

This is definitely not the right fix, you're just removing a symptom
of a much larger problem.

And that problem is that nothing cleans up the objects created by
vnet_new(), they are completely leaked.

vnet_exit(), after doing the vio_unregister_driver(), should call a
helper function that iterates over vnet_list and cleans up those
objects.

That's where the unregister_netdevice() belongs, not in vnet_port_remove().

Re: [PATCH net-next] sunvnet: call unregister_netdev() exactly once per device

[Sowmini Varadhan]

On 07/08/2014 11:42 PM, David Miller wrote:

>
> This is definitely not the right fix, you're just removing a symptom
> of a much larger problem.
>
> And that problem is that nothing cleans up the objects created by
> vnet_new(), they are completely leaked.

the bug(s) in modprobe -r is one manifestation of the problem
and should certainly be fixed, I can look into the suggestions
you have below.

But this is actually a bit worse. Whenever a node goes
away, it triggers a vio_port_remove() on all the peering nodes,
and blows away their networking. IOW this can happen without
involving vnet_exit() at all. Thus it might be necessary
for vio_port_remove() itself to check if its the last port
before doing the unregister?

BTW, for some odd reason I did not see my original patch submission
show up in the list. I'm not sure why.

>
> vnet_exit(), after doing the vio_unregister_driver(), should call a
> helper function that iterates over vnet_list and cleans up those
> objects.
>
> That's where the unregister_netdevice() belongs, not in vnet_port_remove().
> --
> 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
>