Re: [PATCH] CDRW packet writing support for 2.6.7-bk13

[Peter Osterlund <petero2@telia.com>]

Andrew Morton <akpm@osdl.org> writes:

> Greg KH <greg@kroah.com> wrote:
> >
> > > const char *__bdevname(dev_t dev, char *buffer)
> > > {
> > > 	struct gendisk *disk;
> > > 	int part;
> > > 
> > > 	disk = get_gendisk(dev, &part);
> > > 	if (disk) {
> > > 		buffer = disk_name(disk, part, buffer);
> > > 		put_disk(disk);
> > > 	} else {
> > > 		snprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
> > > 				MAJOR(dev), MINOR(dev));
> > > 	}
> > > 
> > > get_gendisk() did an internal module_get() in kobj_lookup().
> > 
> > But if kobj_lookup() succeeds, module_put() is called before returning
> > (actually it's always called, right?) 

But kobj_lookup() also calls probe(), which equals ata_probe() in the
IDE case, and ata_probe() calls get_disk() which calls
try_module_get() to acquire another reference to the module. I guess
all probe() functions behave the same way, as I've seen the same
problem with the sr_mod module.

> Oop, sorry, yes.  Peter, are you sure this is where the leak is coming from?

I'm sure that the module reference count as reported by lsmod
increases each time I access /proc/driver/pktcdvd/pktcdvd0. I can make
this problem go away either by patching __bdevname() or by deleting
the call to __bdevname() in pktcdvd.c.

-- 
Peter Osterlund - petero2@telia.com
http://w1.894.telia.com/~u89404340