On Fri, Apr 24, 2020 at 06:58:23PM -0700, Bart Van Assche wrote: > On 2020-04-24 15:32, Luis Chamberlain wrote: > > On Sun, Apr 19, 2020 at 04:40:45PM -0700, Bart Van Assche wrote: > >> On 4/19/20 12:45 PM, Luis Chamberlain wrote: > >>> Through code inspection I've found that we don't put_device() if > >>> device_add() fails, and this must be done to decrement its refcount. > >> > >> Reviewed-by: Bart Van Assche <bvanassche@xxxxxxx> > > > > Turns out this is wrong, as bdi needs it still, we have can only remove > > it once all users are done, which should be at the disk_release() path. > > > > I've found this while adding the errors paths missing. > > Hi Luis, > > I had a look at the comments above device_add() before I added my > Reviewed-by. Now that I've had another look at these comments and also > at the device_add() implementation I agree that we don't need this patch. Thanks for the confirmation. And just to note, we don't do then put_device() because we don't handle error paths properly. Once we do, we'll need to ensure we put_disk() just at the right place. I'm working on putting some final brush strokes on that now. Luis
