kernel hangs when trying to remove a bridge

kernel hangs when trying to remove a bridge

[Rodrigo Ventura]

I'm using kernel 2.6.16 (-gentoo-r13 actually). A bridge iface br0 is setup 
using eth1 and eth0.1 (VLAN) as slaves. To bring br0, what I do is to remove 
the slaves from the bridge first, "ip link set dev br0 down" next, followed 
by the brctl command to remove the bridge. What happens is that it seems it 
tryes to destroy the bridge iface, and then hangs, with dmesg complaining, 
periodically, about once a second, something like:

unregister_netdevice: waiting for br0 to become free. Usage count = 1
unregister_netdevice: waiting for br0 to become free. Usage count = 1
unregister_netdevice: waiting for br0 to become free. Usage count = 1
...

I say it partially hangs because commands like ifconfig hang (ps state=Disk 
busy)...
To reboot the machine, a HARD reset is required, since, the shutdown process 
hangs... 

Cheers,

Rodrigo

-- 
Rodrigo Ventura
ISR/IST

Re: kernel hangs when trying to remove a bridge

[Stephen Hemminger]

On Sat, 29 Jul 2006 16:42:04 +0100
Rodrigo Ventura <yoda@isr.ist.utl.pt> wrote:

> 
> I'm using kernel 2.6.16 (-gentoo-r13 actually). A bridge iface br0 is setup 
> using eth1 and eth0.1 (VLAN) as slaves. To bring br0, what I do is to remove 
> the slaves from the bridge first, "ip link set dev br0 down" next, followed 
> by the brctl command to remove the bridge. What happens is that it seems it 
> tryes to destroy the bridge iface, and then hangs, with dmesg complaining, 
> periodically, about once a second, something like:
> 
> unregister_netdevice: waiting for br0 to become free. Usage count = 1
> unregister_netdevice: waiting for br0 to become free. Usage count = 1
> unregister_netdevice: waiting for br0 to become free. Usage count = 1
> ...
> 
> I say it partially hangs because commands like ifconfig hang (ps state=Disk 
> busy)...
> To reboot the machine, a HARD reset is required, since, the shutdown process 
> hangs... 
> 

Some broken protocol in the kernel, incremented a refcount but forgot
to cleanup when notified on device removal.

There was a bug in the VLAN code that did that, not sure which version
it was fixed in.

Do you have IPV6 installed (as a module)?
In some kernel versions, IPV6 has a problem with device ref counting, and
leaves a dangling reference.


-- 

Re: kernel hangs when trying to remove a bridge

[Rodrigo Ventura]

On Monday 31 July 2006 19:22, Stephen Hemminger wrote:
> There was a bug in the VLAN code that did that, not sure which version
> it was fixed in.
>
> Do you have IPV6 installed (as a module)?
> In some kernel versions, IPV6 has a problem with device ref counting, and
> leaves a dangling reference.

Yes, I have ipv6 working (dual-stack) and as a module.

The problem I reported was not found in earlier kernels, so it is all working 
fine now.

Cheers,

Rodrigo

-- 
Rodrigo Ventura
ISR/IST