Re: [PATCH 1/2] ublk_drv: move destroying device out of ublk_add_dev

[Ming Lei <ming.lei@redhat.com>]

On Fri, Jul 22, 2022 at 10:58:59AM +0800, Ziyang Zhang wrote:
> On 2022/7/22 10:36, Ming Lei wrote:
> > ublk_device is allocated in ublk_ctrl_add_dev(), so code will become more
> > readable by just letting ublk_ctrl_add_dev() destroy ublk_device in case
> > of ublk_add_dev() failure.
> > 
> > Meantime ub->mutex is destroyed in __ublk_destroy_dev(), but it may
> > not be initialized when ublk_add_dev() fails, so fix it by moving
> > mutex_init(ub->mutex) before any failure path.
> > 
> > Signed-off-by: Ming Lei <ming.lei@redhat.com>
> > ---
> >  drivers/block/ublk_drv.c | 10 +++++-----
> >  1 file changed, 5 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> > index f058f40b639c..d03563286c76 100644
> > --- a/drivers/block/ublk_drv.c
> > +++ b/drivers/block/ublk_drv.c
> > @@ -1106,9 +1106,10 @@ static int ublk_add_dev(struct ublk_device *ub)
> >  
> >  	INIT_WORK(&ub->stop_work, ublk_stop_work_fn);
> >  	INIT_DELAYED_WORK(&ub->monitor_work, ublk_daemon_monitor_work);
> > +	mutex_init(&ub->mutex);
> >  
> >  	if (ublk_init_queues(ub))
> > -		goto out_destroy_dev;
> > +		return err;
> >  
> >  	ub->tag_set.ops = &ublk_mq_ops;
> >  	ub->tag_set.nr_hw_queues = ub->dev_info.nr_hw_queues;
> > @@ -1122,7 +1123,6 @@ static int ublk_add_dev(struct ublk_device *ub)
> >  		goto out_deinit_queues;
> >  
> >  	ublk_align_max_io_size(ub);
> > -	mutex_init(&ub->mutex);
> >  	spin_lock_init(&ub->mm_lock);
> >  
> >  	/* add char dev so that ublksrv daemon can be setup */
> > @@ -1130,8 +1130,6 @@ static int ublk_add_dev(struct ublk_device *ub)
> >  
> >  out_deinit_queues:
> >  	ublk_deinit_queues(ub);
> > -out_destroy_dev:
> > -	__ublk_destroy_dev(ub);
> >  	return err;
> >  }
> >  
> > @@ -1331,8 +1329,10 @@ static int ublk_ctrl_add_dev(struct io_uring_cmd *cmd)
> >  	ub->dev_info.dev_id = ub->ub_number;
> >  
> >  	ret = ublk_add_dev(ub);
> > -	if (ret)
> > +	if (ret) {
> > +		__ublk_destroy_dev(ub);
> >  		goto out_unlock;
> > +	}
> 
> Hi, Ming.
> 
> Now, if ublk_add_dev() returns failure, __ublk_destroy_dev() is called anyway.
> 
> However, in current ublk_drv:ublk_add_dev():
> 
> ...
> 	return ublk_add_chdev(ub);   <---- here
> out_deinit_queues:
> 	ublk_deinit_queues(ub);
> out_destroy_dev:
> 	__ublk_destroy_dev(ub);
> 	return err;
> 
> 
> ublk_add_chdev() returns and the returned value(maybe a failure) directly
> pass to ublk_ctrl_add_dev which does NOT call __ublk_destroy_dev()
> 
> please check it is correct to call __ublk_destroy_dev() if ublk_add_chdev() fails.

If ublk_add_chdev fails, we shouldn't call __ublk_destroy_dev() any
more, since ublk_add_chdev() does handle the cleanup, so this patch
is wrong.

will fix it in V2.


Thanks,
Ming