Hello, On Thu, Oct 31, 2024 at 04:04:00PM +0800, Yu Kuai wrote: > > On Mon, Sep 30, 2024 at 04:52:58PM +0800, Yu Kuai wrote: > > > +static inline bool blkg_print_dev_name(struct seq_file *sf, > > > + struct blkcg_gq *blkg) > > > +{ > > > + struct gendisk *disk = blkg->q->disk; > > > + > > > + if (!disk) > > > + return false; > > > + > > > + seq_printf(sf, "%u:%u", disk->major, disk->first_minor); > > > + return true; > > > +} > > > + > > > > I wonder whether we just should add a name field to disk. > > And suggestions on this set now? I guess add a name filed is not > appropriate. :( Yeah, I don't know. I've always struggled a bit with block device names. We use MAJ:MIN in all the input interfaces and mix the disk names and MAJ:MIN when outputting and there are (or is it used to be now?) request_queues without disk attached, so sometimes names are just not available. Jens, do you any preference here? The proposed patch can be fine but e.g. it can race against disk_release() if the caller isn't careful and it also sucks not knowing the name in destruction path. Thanks. -- tejun