Re: [RFC] fix sysfs symlinks

[Rusty Russell <rusty@rustcorp.com.au>]

On Fri, 2004-04-16 at 05:14, viro@parcelfarce.linux.theplanet.co.uk
wrote:
> BTW, how about a new section that would
> 	a) be allocated separately at module load time
> 	b) contain a kobject with ->release() freeing that section
> 	c) be populated with structures containing kobjects and having
> no ->release(); main kobject would be pinned down by them.  Original
> refcount in each of those guys would be 1.
> 
> module_exit() would unregister all stuff we have in there and then drop
> the references to them.  No waiting for anything and when all references
> to these objects are gone, we get the section freed.  That can happen
> way after the completion of rmmod - as the matter of fact we could have
> the same module loaded again by that time.

Or you could skip the extra section, and keep all the module memory
until later.  Instead of a section marker, you then set the release of
those static things to "static_release" which does the put on the module
memory kref:

void static_release(struct kobject *kobj)
{
	struct module *mod;

	down(&module_mutex);
	list_for_each_entry(mod, &modules, list) {
		BUG_ON(within(kobj, mod->module_init, mod->init_size);
		if (within(kobj, mod->module_core, mod->core_size)) {
			kref_put(&mod->mem_kref);
			up(&module_mutex);
			return;
		}
	}
	up(&module_mutex);
	BUG();
}

One question which comes to mind, who does the original kref_get()s on
the module memory?  Even if we did have a separate section, the module
loader can't know how many objects objects are in there...

Cheers,
Rusty.
-- 
Anyone who quotes me in their signature is an idiot -- Rusty Russell