Re: [RFC PATCH 8/9] livepatch: allow patch modules to be removed

[Jiri Slaby <jslaby@suse.cz>]

On 02/10/2015, 08:57 PM, Josh Poimboeuf wrote:
> On Tue, Feb 10, 2015 at 08:02:34PM +0100, Jiri Slaby wrote:
>> On 02/09/2015, 06:31 PM, Josh Poimboeuf wrote:
>>> --- a/kernel/livepatch/core.c
>>> +++ b/kernel/livepatch/core.c
>> ...
>>> @@ -497,10 +500,6 @@ static struct attribute *klp_patch_attrs[] = {
>>>  
>>>  static void klp_kobj_release_patch(struct kobject *kobj)
>>>  {
>>> -	/*
>>> -	 * Once we have a consistency model we'll need to module_put() the
>>> -	 * patch module here.  See klp_register_patch() for more details.
>>> -	 */
>>
>> I deliberately let you write the note in there :). What happens when I
>> leave some attribute in /sys open and you remove the module in the meantime?
> 
> You're right, as was I the first time :-)
> 
> The only problem is that it would be nice if we could call
> klp_unregister_patch() from the patch module's exit function, so that
> doing an rmmod on the patch module unregisters it.  But if we put
> module_put() in the patch release function, then we have a circular
> dependency and we could never rmmod it.
>
> How about instead we do a klp_is_patch_registered() at the beginning of
> all the attribute accessor functions?  It's kind of ugly, but I can't
> think of a better idea at the moment.

Ugh, no :). You even have the kobject proper in the module which would
be gone.

However we can take inspiration in kgraft. I introduced a completion
there and wait for it in rmmod. This completion is made complete in
kobject's release. See:
https://git.kernel.org/cgit/linux/kernel/git/jirislaby/kgraft.git/tree/kernel/kgraft_files.c?h=kgraft#n30
https://git.kernel.org/cgit/linux/kernel/git/jirislaby/kgraft.git/tree/kernel/kgraft_files.c?h=kgraft#n138

This should IMO work here too.

regards,
-- 
js
suse labs