Re: [PATCH] RDMA: don't ignore client->add() failures

[Jason Gunthorpe <jgg@ziepe.ca>]

On Thu, Mar 30, 2023 at 08:51:44AM +0900, Tetsuo Handa wrote:
> On 2023/03/30 0:44, Jason Gunthorpe wrote:
> >> The caller of ib_register_device() (i.e. siw_device_register() from
> >> siw_newlink()) is assuming that somebody will call __ib_unregister_device(),
> >> but nobody is calling __ib_unregister_device().
> > 
> > On the success path this stuff happens during dellink
> > 
> 
> "struct rdma_link_ops" has "newlink" callback but does not have "dellink" callback.
> "struct rtnl_link_ops" has both "newlink" callback and "dellink" callback, but
> only ipoib_netlink.c defines it inside drivers/infiniband/ directory.
> 
> Where is the dellink you are talking about?
> I'm not familiar with rdma code. Please explain using specific function/symbol names.

Your test code deletes a vlan

If SIW is attached to a vlan then SIW will destroy itself based on a
net notifier

Look at siw_netdev_event:

	case NETDEV_UNREGISTER:
		ib_unregister_device_queued(&sdev->base_dev);
		break;

Jason