netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* unregister_netdev() deadlock
@ 2008-02-27 15:44 Jeba Anandhan
  2008-02-27 16:46 ` Stephen Hemminger
  0 siblings, 1 reply; 3+ messages in thread
From: Jeba Anandhan @ 2008-02-27 15:44 UTC (permalink / raw)
  To: netdev; +Cc: matthew.hattersley

Hi all,
I have doubts about unregister the netdevice.

Scenario
I have two netdevice entities. I wish to unregister one entity.

struct net_device *dev1;
struct net_device *dev2;

dev1->timer = XXX;
dev2->timer = YYY;

dev1->ioctl() {

...
case delete_other:
            unregister_netdev(dev2);
            free_netdev(dev2);
            return 0;
...
}

The dev2->refcnt is not zero in this case. unregister_netdev(dev2) cause
deadlock. Could you tell me why it happens?.


It works smoothly when we call the unregister_netdev(dev2) from
cleanup_module(). dev2->refcnt is not zero in this case.  Still it is
able to unregister.


Thanks
Jeba



^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: unregister_netdev() deadlock
  2008-02-27 15:44 unregister_netdev() deadlock Jeba Anandhan
@ 2008-02-27 16:46 ` Stephen Hemminger
  2008-02-27 17:05   ` Jeba Anandhan
  0 siblings, 1 reply; 3+ messages in thread
From: Stephen Hemminger @ 2008-02-27 16:46 UTC (permalink / raw)
  To: Jeba Anandhan; +Cc: netdev, matthew.hattersley

On Wed, 27 Feb 2008 15:44:54 +0000
Jeba Anandhan <jeba.anandhan@vaioni.com> wrote:

> Hi all,
> I have doubts about unregister the netdevice.
> 
> Scenario
> I have two netdevice entities. I wish to unregister one entity.
> 
> struct net_device *dev1;
> struct net_device *dev2;
> 
> dev1->timer = XXX;
> dev2->timer = YYY;
> 
> dev1->ioctl() {
> 
> ...
> case delete_other:
>             unregister_netdev(dev2);
>             free_netdev(dev2);
>             return 0;
> ...
> }
> 
> The dev2->refcnt is not zero in this case. unregister_netdev(dev2) cause
> deadlock. Could you tell me why it happens?.
> 
> 
> It works smoothly when we call the unregister_netdev(dev2) from
> cleanup_module(). dev2->refcnt is not zero in this case.  Still it is
> able to unregister.
> 
> 
> Thanks
> Jeba
> 

You need to properly manage device refcounts (or delete both).
Also since dev1->ioctl is called with RTNL you can't:
    use unregister_netdev() recursive locking (use unregister_netdevice)
    or  call free_netdev because of netdevice could still be in use (use dev->destructor instead).

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: unregister_netdev() deadlock
  2008-02-27 16:46 ` Stephen Hemminger
@ 2008-02-27 17:05   ` Jeba Anandhan
  0 siblings, 0 replies; 3+ messages in thread
From: Jeba Anandhan @ 2008-02-27 17:05 UTC (permalink / raw)
  To: Stephen Hemminger; +Cc: netdev, matthew.hattersley

On Wed, 2008-02-27 at 08:46 -0800, Stephen Hemminger wrote:
> On Wed, 27 Feb 2008 15:44:54 +0000
> Jeba Anandhan <jeba.anandhan@vaioni.com> wrote:
> 
> > Hi all,
> > I have doubts about unregister the netdevice.
> > 
> > Scenario
> > I have two netdevice entities. I wish to unregister one entity.
> > 
> > struct net_device *dev1;
> > struct net_device *dev2;
> > 
> > dev1->timer = XXX;
> > dev2->timer = YYY;
> > 
> > dev1->ioctl() {
> > 
> > ...
> > case delete_other:
> >             unregister_netdev(dev2);
> >             free_netdev(dev2);
> >             return 0;
> > ...
> > }
> > 
> > The dev2->refcnt is not zero in this case. unregister_netdev(dev2) cause
> > deadlock. Could you tell me why it happens?.
> > 
> > 
> > It works smoothly when we call the unregister_netdev(dev2) from
> > cleanup_module(). dev2->refcnt is not zero in this case.  Still it is
> > able to unregister.
> > 
> > 
> > Thanks
> > Jeba
> > 
> 
> You need to properly manage device refcounts (or delete both).
> Also since dev1->ioctl is called with RTNL you can't:
>     use unregister_netdev() recursive locking (use unregister_netdevice)
>     or  call free_netdev because of netdevice could still be in use (use dev->destructor instead).

*) Is there any chance of deadlock if i try to unregister_netdev() on
the fly using syscall?
*) is dev->refcnt altered by only dev_hold or dev_put?. what are the
general cases where the dev->refcnt is altered?

Thanks
Jeba

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2008-02-27 17:07 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2008-02-27 15:44 unregister_netdev() deadlock Jeba Anandhan
2008-02-27 16:46 ` Stephen Hemminger
2008-02-27 17:05   ` Jeba Anandhan

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).