Re: [PATCH][ATM] use rtnl_{lock,unlock} during device operations

[Roman Zippel]

Hi,

On Sat, 7 Jun 2003, David S. Miller wrote:

>    The basic problem is still that module_exit is a synchronous interface, 
>    from where you can't call any asynchronous functions, unless you prevent 
>    new callbacks via try_module_get() or you have to wait for all pending 
>    events.
> 
> We don't use module refcounts at all anymore for netdev objects.
> That's the whole key.

If I read the source correctly, unregister_netdevice() simply does rip the 
device out and sees what happens? It's a bit primitive, but should of 
course work too.
I'm just not sure if that model works that well in other parts too, I 
think file systems don't really like it, if the device under them 
disappears. Any kind of automatic module unloading also becomes a bit 
difficult without a module use count. I also hope no net driver starts 
exporting it's own data via sysfs (I currently don't see anything that 
would protect this).
BTW does anything protect the get_stats() call in netstat_attr_show()?

bye, Roman