Re: [RFC] New kobject/kset/ktype documentation and example code

[Cornelia Huck <cornelia.huck@de.ibm.com>]

On Wed, 28 Nov 2007 14:18:18 -0500 (EST),
Alan Stern <stern@rowland.harvard.edu> wrote:

> On Wed, 28 Nov 2007, Cornelia Huck wrote:
> 
> > We should perhaps add a bit fat warning here:
> > 
> > Note that once you registered your kobject via kobject_add(), you must
> > never use kfree() to free it directly. The only safe way is to use
> 
> Slightly ambiguous.  Instead just say:
> 
> If you have initialized your kobject via kobject_init() or
> kobject_register(), you must not deallocate the kobject anywhere other
> than its release() method (which is invoked during the final
> kobject_put() call).  Otherwise the kernel will leak memory.

And references to the kobject's kset...

> 
> > > One important point cannot be overstated: every kobject must have a
> > > release() method, and the kobject must persist (in a consistent state)
> > > until that method is called. 
> > 
> > Which is especially hurting if you use kobjects in modules. (Which
> > reminds me: Must dig up the patchset that fixes the module unload vs.
> > release problem.)
> 
> In theory modules shouldn't present a problem -- especially if Greg 
> merges the "Kobjects: drop child->parent ref at unregistration" patch.
> 
> When a module is unloaded, it has to unregister all its kobjects, which 
> should force all their children to be unregistered too.  At that time 
> the children's drivers should drop all their references to the parent 
> kobject, leaving only references held by the module being unloaded.  
> Presumably it can arrange to drop its own references before its exit() 
> routine returns.
> 
> The only problem arises when a child's driver retains a reference to 
> the parent kobject.  If things are done properly, this reference should 
> involve incrementing the module count -- which would prevent the module 
> from being unloaded in the first place.

This still leaves the possibility that random code may grab a reference
once the kobject is present in the tree and lookupable. You gave up the
control of the number of references to your object once you made it
public.