Re: [PATCH] kobject: make sure kobj->ktype is set before kobject_init

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

On Thu, 29 Nov 2007 11:59:06 +0100,
Kay Sievers <kay.sievers@vrfy.org> wrote:

> On Thu, 2007-11-29 at 11:05 +0100, Cornelia Huck wrote:
> > On Wed, 28 Nov 2007 17:00:57 -0500 (EST),
> > Alan Stern <stern@rowland.harvard.edu> wrote:
> > > On Wed, 28 Nov 2007, Greg KH wrote:
> > > > On Wed, Nov 28, 2007 at 03:42:00PM -0500, Alan Stern wrote:
> > > > > This patch (as1020) adds a check to kobject_init() to insure that the
> > > > > ktype field is not NULL.  This is just for safety's sake; as far as I
> > > > > know there are no remaining places where the field is left unset.  But
> > > > > ironically, kset_init() did fail to set it!  The patch fixes that and
> > > > > removes some redundant initialization in kset_createa().
> > > > > 
> > > > > The patch also fixes up elevator_init(), where ktype was set after
> > > > > calling kobject_init() instead of before.
> > > > 
> > > > No, it's safe to set ktype after kobject_init, it was just not safe to
> > > > set the kset.  As Kay pointed out, I just added a patch to my tree to
> > > > resolve this issue, and I'll go back and update the documentation now.
> > > > 
> > > > I do like the "check for a ktype" warning, I'll go add that to my local
> > > > tree and see what breaks.  Odds are, all the static kobjects will :(
> > > 
> > > You have to be careful.  The ktype check I wrote lives in
> > > kobject_init() -- that's why I had to make elevator_init() assign the
> > > ktype before calling kobject_init().  If you put the check into
> > > kobject_add() instead then you won't end up checking objects that get
> > > initialized but not added.
> > > 
> > > Yes, nobody would deliberately use a kobject without adding it, but it 
> > > could happen as the result of an failure between the _init and _add 
> > > calls.
> > 
> > And if someone calls kobject_put() after kobject_init() to clean up,
> > their release function will not be called if they didn't set the ktype.
> > So the check really belongs into kobject_init() IMO.
> 
> Hmm, will one expect that the whole object will also be free'd when we
> suggest to call kobject_put() to cleanup? That might be pretty
> unexpected, right?

I'd expect the kobject to be freed if I called kobject_put() on my last
reference (but that may be because I'm familiar with the code :)).
OTOH, if getting the reference on the kset is moved from kobject_init()
to kobject_add(), we aren't forced to use kobject_put() but may use
kfree().
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/