On Tue 14-04-20 08:52:28, Christoph Hellwig wrote: > Looking through this series the whoe approach of using a lock to clear > the ->dev pointer looks rather odd to me. What is the reason for now > simply adding a separately allocated name field to struct > backing_dev_info that the name is copied to on allocation, and then > the ->dev field is not relevant for name printing and we don't need > to copy out the name in the potentionally more fast path callers that > want to print it? Yeah, that's what I was suggesting as well [1] - especially since we already have bdi->name with a dubious value (but looking into it now, we would need a separate dev_name field since bdi->name is visible in sysfs so we cannot change that). But Yufen explained to me that this could result in bogus name being reported when bdi gets re-registered. Not sure if that's serious enough but it could happen... Honza [1] https://lore.kernel.org/linux-block/20200219125505.GP16121@xxxxxxxxxxxxxx -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
