> That ioctl interface is just plain horrible, it should really stay out > of view inside libdevmapper in case somebody notices. And we do not > want to make libdevmapper a dependency just to ioctl the virtual > device. Going through libdevmapper makes plenty of sense for things > like suspend/resume where there is a lot of deadlock-prevention logic > that has to be followed rigorously, but no sense whatsoever for what I > use this interface, and what I expect other people might use it for. > Of course, you are one of those "other people", I think, with > multipath, so it would help if you tell me what you have in mind there. > Do you volunteer to resurrect the dmfs pseudo-fs ? :)